Partially Relevant Video Retrieval (PRVR) aims to retrieve untrimmed videos containing relevant moments for a given text query. This task is extremely challenging, as untrimmed videos often include numerous actions and objects unrelated to the query. However, existing methods usually struggle with fine-grained action-object modeling, limiting their retrieval performance. To tackle this challenge, we introduce Action-and-object Aware Alignment for Partially Relevant Video Retrieval (A3PRVR), a dual-branch framework designed to enhance retrieval by improving the modeling of action-object relationships. Specifically, we propose a Query-specific Deformable Temporal Attention (Q-DTA) module to effectively capture action-relevant object information in video features, while filtering out irrelevant content. Additionally, we propose an action-and-object aware alignment module to enable fine-grained textual understanding and video-text alignment. It uses action- and object-aware contrastive losses to enhance the model's sensitivity to action-object distinctions in the text query. Compared to state-of-the-art methods, A3PRVR achieves an average relative gain of 6.5% in SumR across the Charades-STA, ActivityNet-Caption, and TVR datasets.

Embodied navigation is a fundamental capability for intelligent agents, yet remains challenging in partially observable environments where navigation instructions can be difficult to interpret. However, existing tasks only provide unimodal instructions, which are ambiguous in complex multimodal environments with multiple similar objects, and may result in misinterpretation and navigation failure. To overcome these limitations, we introduce MINav, a novel task where the navigation path is precisely described by a multimodal instruction. The instruction provides multimodal cues, including object categories, RGB images, language descriptions, and auditory descriptions, which help the agent to disambiguate and ground objects in the environment and navigate effectively. We further construct a large-scale dataset of 43.9K navigation episodes using a two-stage pipeline that first annotates multimodal references of objects and then synthesizes diverse multimodal instructions. We find that existing methods struggle on MINav task, indicating substantial room for improvement in agents' multimodal grounding. To address this, we propose NaVLA^2, a vision-language-audio-action model that additionally integrates spatial audio and employs a CoThinkAct module to jointly generate high-level reasoning and consistent low-level actions. Experimental results demonstrate that NaVLA^2 significantly outperforms competitive baselines on MINav benchmark. We hope that our proposed MINav and NaVLA^2 will facilitate future research toward agents with stronger multimodal understanding and grounding capabilities for navigation.

Learning multimodal representation is a fundamental task that supports a wide range of applications such as visual-text retrieval. While pioneering approaches e.g., CLIP paves the way by learning separated encoders for different modalities, they struggle to model complex interactions between modalities, resulting in inferior vision and language representation. Recently, researchers have begun to leverage powerful Large Vision-Language Models (LVLMs) for unimodal or multimodal encoding, showing substantial improvement over separated encoder methods. However, we find that directly adapting LVLMs to embedding models suffers from insufficient visual representation and coarse multimodal alignment. To address these issues, we propose a simple yet effective Fine-grained Alignment Matters (FAM) method to achieve fine-grained vision-language embedding learning with LVLMs. First, to close the gap between the pure generation and multimodal embedding using LVLMs, we propose Multi-granularity Aligned Contrastive (MAC) to explicitly learn and align fine-grained modality representations at multiple granularity levels using image-text pairs. Second, to mitigate the insufficiency of visual representation during adapting LVLMs to downstream embedding tasks, we propose a Vision Embedding Inversion Training (VEIN) strategy to encourage the extracted embeddings to preserve fine-grained visual features. Extensive experiments demonstrate the effectiveness of our method, which achieves superior performance on various downstream multimodal datasets.

Large Language Models (LLMs) hold rich implicit knowledge and powerful transferability. In this paper, we explore the combination of LLMs with the human skeleton to perform action classification and description. However, when treating LLM as a recognizer, two questions arise: 1) How can LLMs understand the skeleton? 2) How can LLMs distinguish among actions? To address these problems, we introduce a novel paradigm named learning Skeleton representation with visual-motion knowledge for Action Recognition (SUGAR). In our pipeline, we first utilize off-the-shelf large-scale video models as a knowledge base to generate visual, motion information related to actions. Then, we propose to supervise skeleton learning through this prior knowledge to yield discrete representations. Finally, we use the LLM with untouched pre-training weights to understand these representations and generate the desired action targets and descriptions. Notably, we present a Temporal Query Projection (TQP) module to continuously model the skeleton signals with long sequences. Experiments on several skeleton-based action classification benchmarks demonstrate the efficacy of our SUGAR. Moreover, experiments on zero-shot scenarios show that SUGAR is more versatile than linear-based methods.

Zeng You, Yaofo Chen, Qiuwu Chen, Ying Sun, Shuhai Zhang, Yingjian Li, Yaowei Wang, Mingkui Tan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zero-shot temporal action localization (ZSTAL) aims to localize and recognize action categories unseen during training. However, it assumes that test videos contain only unseen classes, which is unrealistic in practice where seen and unseen actions naturally co-exist. To bridge this gap, we introduce generalized ZSTAL (GZS-TAL), where models trained only on seen classes must handle both seen and unseen ones during testing. This setting highlights a critical challenge: a static, frozen model cannot adapt to the mixed distributions encountered at test time. To address this issue, we propose a Temporal-Sensitive Adaptation (TSA) module that equips TAL models with the ability to update themselves during testing. The key intuition is to use temporal dependency prediction as a self-supervised signal: TSA introduces an online-updatable memory optimized to reconstruct features of preceding segments from the current one, thereby embedding temporal dependencies into parameters and reusing them for adaptation at test time. To further enhance temporal modeling, we extend TSA into a Bi-directional TSA (Bi-TSA) mechanism that performs prediction in both forward and backward directions. By simultaneously exploiting historical and future contexts, Bi-TSA improves long-range temporal representation and yields more accurate boundary localization. Extensive experiments on THUMOS14 and ActivityNet-1.3 demonstrate that our approach achieves significant improvements over state-of-the-art methods under the GZS-TAL setting, validating its effectiveness and generalization ability.

Large language models (LLMs) have shown outstanding performance across a variety of tasks, partly due to advanced prompting techniques. However, these techniques often require lengthy prompts, which increase computational costs and can hinder performance because of the limited context windows of LLMs. While prompt compression is a straightforward solution, existing methods confront the challenges of retaining essential information, adapting to context changes, and remaining effective across different tasks. To tackle these issues, we propose a task-agnostic method called Dynamic Prompt Compression (LLM-DPC). Our method reduces the number of prompt tokens while minimizing any degradation in LLM performance. We model prompt compression as a Markov Decision Process (MDP), enabling the DPC-Agent to sequentially remove redundant tokens by adapting to dynamic contexts and retaining crucial content. We develop a reward function for training the DPC-Agent that balances the compression ratio, the quality of the LLM output, and the retention of key information. This allows for prompt token reduction without needing an external black-box LLM. Inspired by the progressive difficulty adjustment in curriculum learning, we introduce a Hierarchical Prompt Compression (HPC) training strategy that gradually increases the compression difficulty, enabling the DPC-Agent to learn an effective compression method that maintains information integrity. Experiments demonstrate that our method outperforms state-of-the-art techniques, especially at higher compression ratio.

Emotion recognition based on body movements is vital in human-computer interaction. However, existing emotion recognition methods predominantly focus on enhancing classification accuracy, often neglecting the provision of textual explanations to justify their classifications. In this paper, we propose an Emotion-Action Interpreter powered by LargeLanguage Model (EAI-LLM), which not only recognizes emotions but also generates textual explanations by treating 3D body movement data as unique input tokens within large language models (LLMs). Specifically, we propose a multi-granularity skeleton tokenizer designed for LLMs, which separately extracts spatio-temporal tokens and semantic tokens from the skeleton data. This approach allows LLMs to generate more nuanced classification descriptions while maintaining robust classification performance. Furthermore, we treat the skeleton sequence as a specific language and propose a unified skeleton token module. This module leverages the extensive background knowledge and language processing capabilities of LLMs to address the challenges of joint training on heterogeneous datasets, thereby significantly enhancing recognition accuracy on individual datasets. Experimental results demonstrate that our model achieves recognition accuracy comparable to existing methods. More importantly, with the support of background knowledge from LLMs, our model can generate detailed emotion descriptions based on classification results, even when trained on a limited amount of labeled skeleton data.

Large Language Models (LLMs) have achieved remarkable success but face significant deployment challenges in cloud and edge environments due to their massive computational and storage requirements. Model quantization serves as a key solution to enhance the scalability and efficiency of LLMs within distributed cloud platforms. Existing Post-Training Quantization (PTQ) methods often exhibit suboptimal performance in low-bit settings. To further improve their precision, Quantization-Aware Training (QAT) combined with Low-Rank Adaptation (LoRA) has been explored for error correction. However, a critical issue is that the quantized base model and full-precision LoRA parameters suffer from precision mismatch, introducing additional errors during weight merging. To address these challenges, we propose a Quantized Low-rank Error Reconstructor (QER) for LLM low-bitwidth quantization. QER first enables lossless merging in low-bitwidth format by aligning the bitwidth of its low-rank parameters with the quantized base parameters, eliminating dequantization and requantization steps. Through this process, QER reconstructs original errors into two components: the quantization errors of QER parameters (i.e., quantized low-rank parameters) and potential overflow errors during low-bitwidth merging. These two errors are directly related to QER parameters, making them easier to optimize via gradient-based updates within an error-aware training framework. Requiring only 128 samples and 1 training epoch, QER demonstrates superior performance on LLaMA-1/2 families. In 4-bit quantization, compared to QLLM with error correction, QER reduces average perplexity by 13.8% (from 10.97 to 9.45) and improves average accuracy by 3.01 percentage points (from 51.84% to 54.85%) on LLaMA-1-7B. QER bridges the gap between quantization and low-rank adaptation, enabling efficient and accurate low-precision LLM deployment.

Research on 3D Vision-Language Models (3D-VLMs) is gaining increasing attention, which is crucial for developing embodied AI within 3D scenes, such as visual navigation and embodied question answering. Due to the high density of visual features, especially in large 3D scenes, accurately locating task-relevant visual information is challenging. Existing works attempt to segment all objects and consider their features as scene representations. However, these task-agnostic object features include much redundant information and missing details for the task-relevant area. To tackle these problems, we propose LSceneLLM, an adaptive framework that automatically identifies task-relevant areas by leveraging LLM’s visual preference for different tasks, followed by a plug-and-play scene magnifier module to capture fine-grained details in focused areas. Specifically, a dense token selector examines the attention map of LLM to identify visual preferences for the instruction input. It then magnifies fine-grained details of the focusing area. An adaptive self-attention module is leveraged to fuse the coarse-grained and selected fine-grained visual information. To comprehensively evaluate the large scene understanding ability of 3D-VLMs, we further introduce a cross-room understanding benchmark, XR-Scene, which contains a series of large scene understanding tasks including XR-QA, XR-EmbodiedPlanning, and XR-SceneCaption. Experiments show that our method surpasses existing methods on both large scene understanding and existing scene understanding benchmarks. Plunging our scene magnifier module into the existing 3D-VLMs also brings significant improvement. Code and data are available at https://github.com/Hoyyyaard/LSceneLLM

Visual autoregressive modeling, based on the next-scale prediction paradigm, exhibits notable advantages in image quality and model scalability over traditional autoregressive and diffusion models. It generates images by progressively refining resolution across multiple stages. However, the computational overhead in high-resolution stages remains a critical challenge due to the substantial number of tokens involved. In this paper, we introduce SparseVAR, a plug-and-play acceleration framework for next-scale prediction that dynamically excludes low-frequency tokens during inference without requiring additional training. Our approach is motivated by the observation that tokens in low-frequency regions have a negligible impact on image quality in high-resolution stages and exhibit strong similarity with neighboring tokens. Additionally, we observe that different blocks in the next-scale prediction model focus on distinct regions, with some concentrating on high-frequency areas. SparseVAR leverages these insights by employing lightweight MSE-based metrics to identify low-frequency tokens while preserving the fidelity of excluded regions through a small set of uniformly sampled anchor tokens. By significantly reducing the computational cost while maintaining high image generation quality, SparseVAR achieves notable acceleration in both HART and Infinity. Specifically, SparseVAR achieves up to a 2 times speedup with minimal quality degradation in Infinity-2B.

Creating a realistic clothed human from a single-view RGB image is crucial for applications like mixed reality and filmmaking. Despite some progress in recent years, mainstream methods often fail to fully utilize side-view information, as the input single-view image contains front-view information only. This leads to globally unrealistic topology and local surface inconsistency in side views. To address these, we introduce Clothed Human Reconstruction with Side View Consistency, namely CHRIS, which consists of 1) A Side-View Normal Discriminator that enhances global visual reasonability by distinguishing the generated side-view normals from the ground truth ones; 2) A Multi-to-One Gradient Computation (M2O) that ensures local surface consistency. M2O calculates the gradient of a sampling point by integrating the gradients of the nearby points, effectively acting as a smooth operation. Experimental results demonstrate that CHRIS achieves state-of-the-art performance on public benchmarks and outperforms the prior work.

Transformer-based Large Language Models (LLMs) have exhibited remarkable success in extensive tasks primarily attributed to self-attention mechanism, which requires a token to consider all preceding tokens as its context to compute attention. However, when the context length L becomes very large (e.g., 128K), the amount of potentially redundant information in the context tends to increase. The redundant context not only hampers the modeling representation performance but also incurs unnecessary computational and storage overhead. In this paper, we propose a plug-and-play Core Context Aware (CCA) Attention for efficient long-context modeling, comprising two complementary modules: 1) Globality-aware pooling module groups input tokens and dynamically compresses each group into one core token based on their significance. In this way, our method automatically focuses and strengthens core context while diminishing redundancy during the learning process, leading to effective long-term dependency modeling. 2) Locality-preserving module incorporates neighboring tokens to preserve local context for detailed representation. Notably, our CCA-Attention is able to replace the self-attention module in existing LLMs with minimal fine-tuning cost. Extensive experimental results show the superiority of our method in both long-context modeling and computational efficiency over state-of-the-art methods.

While Large Language Models (LLMs) have exhibited remarkable emergent capabilities through extensive pre-training, they still face critical limitations in generalizing to specialized domains and handling diverse linguistic variations, known as distribution shifts. In this paper, we propose a Test-Time Learning (TTL) paradigm for LLMs, namely TLM, which dynamically adapts LLMs to target domains using only unlabeled test data during testing. Specifically, we first provide empirical evidence and theoretical insights to reveal that more accurate predictions from LLMs can be achieved by minimizing the input perplexity of the unlabeled test data. Based on this insight, we formulate the Test-Time Learning process of LLMs as input perplexity minimization, enabling self-supervised enhancement of LLM performance. Furthermore, we observe that high-perplexity samples tend to be more informative for model optimization. Accordingly, we introduce a Sample Efficient Learning Strategy that actively selects and emphasizes these high-perplexity samples for test-time updates. Lastly, to mitigate catastrophic forgetting and ensure adaptation stability, we adopt Low-Rank Adaptation (LoRA) instead of full-parameter optimization, which allows lightweight model updates while preserving more original knowledge from the model. We introduce the AdaptEval benchmark for TTL and demonstrate through experiments that TLM improves performance by at least 20% compared to original LLMs on domain knowledge adaptation.

Offline reinforcement learning is widely applied in multiple fields due to its advantages in efficiency and risk control. However, a major problem it faces is the distribution shift between offline datasets and online environments. This mismatch leads to out-of-distribution (OOD) state-action pairs that fall outside the scope of the training data. Therefore, existing conservative training policies may not provide reliable decisions when the test environment deviates greatly from the offline dataset. In this paper, we propose Test-time Adapted Reinforcement Learning (TARL) to address this problem. TARL constructs unsupervised test-time optimization objectives for discrete and continuous control tasks, using test data without depending on environmental rewards. In discrete control tasks, it minimizes the entropy of predicted action probabilities to decrease uncertainty and avoid OOD state-action pairs. For continuous control tasks, it represents and minimizes action uncertainty based on the normal distribution of policy network outputs. Moreover, to prevent model bias caused by overfitting and error accumulation during the test-time update process, TARL enforces a KL divergence constraint between the fine-tuned policy and the original policy. For efficiency, TARL only updates the layer normalization layer parameters during testing. Extensive experiments on popular Atari game benchmarks and the D4RL dataset demonstrate the superiority of our method. Our method achieved a significant improvement over CQL, with a 13.6% episode return relative increase on the hopper-expert-v2 task.

Transformer-based large language models (LLMs) excel in natural language processing tasks by capturing long-range dependencies through self-attention mechanisms. However, long-context modeling faces significant computational inefficiencies due to redundant attention computations: while attention weights are often sparse, all tokens consume equal computational resources. In this paper, we reformulate traditional probabilistic sequence modeling as a supervised learning task, enabling the separation of relevant and irrelevant tokens and providing a clearer understanding of redundancy. Based on this reformulation, we theoretically analyze attention sparsity, revealing that only a few tokens significantly contribute to predictions. Building on this, we formulate attention optimization as a linear coding problem and propose a group coding strategy, theoretically showing its ability to improve robustness against random noise and enhance learning efficiency. Motivated by this, we propose Dynamic Group Attention (DGA), which leverages the group coding to explicitly reduce redundancy by aggregating less important tokens during attention computation. Empirical results show that our DGA significantly reduces computational costs while maintaining competitive performance.

Electromagnetic structure (EMS) design plays a critical role in developing advanced antennas and materials, but remains challenging due to high-dimensional design spaces and expensive evaluations. While existing methods commonly employ high-quality predictors or generators to alleviate evaluations, they are often data-intensive and struggle with real-world scale and budget constraints. To address this, we propose a novel method called Progressive Quadtree-based Search (PQS). Rather than exhaustively exploring the high-dimensional space, PQS converts the conventional image-like layout into a quadtree-based hierarchical representation, enabling a progressive search from global patterns to local details. Furthermore, to lessen reliance on highly accurate predictors, we introduce a consistency-driven sample selection mechanism. This mechanism quantifies the reliability of predictions, balancing exploitation and exploration when selecting candidate designs. We evaluate PQS on two real-world engineering tasks, i.e., Dual-layer Frequency Selective Surface and High-gain Antenna. Experimental results show that our method can achieve satisfactory designs under limited computational budgets, outperforming baseline methods. In particular, compared to generative approaches, it cuts evaluation costs by 75$\tilde$85%, effectively saving 20.27$\tilde$38.80 days of product designing cycle.

Vision-and-Language Navigation (VLN) tasks require an agent to follow textual instructions to navigate through 3D environments. Traditional approaches use supervised learning methods, relying heavily on domain-specific datasets to train VLN models. Recent methods try to utilize closedsource large language models (LLMs) like GPT-4 to solve VLN tasks in zero-shot manners, but face challenges related to expensive token costs and potential data breaches in realworld applications. In this work, we introduce Open-Nav, a novel study that explores open-source LLMs for zero-shot VLN in the continuous environment. Open-Nav employs a spatial-temporal chain-of-thought (CoT) reasoning approach to break down tasks into instruction comprehension, progress estimation, and decision-making. It enhances scene perceptions with fine-grained object and spatial knowledge to improve LLM's reasoning in navigation. Our extensive experiments in both simulated and real-world environments demonstrate that Open-Nav achieves competitive performance compared to using closed-source LLMs.

Open-domain dialogue systems aim to generate natural and engaging conversations, providing significant practical value in real applications such as social robotics and personal assistants. The advent of large language models (LLMs) has greatly advanced this field by improving context understanding and conversational fluency. However, existing LLM-based dialogue systems often fall short in proactively understanding the user's chatting preferences and guiding conversations toward user-centered topics. This lack of user-oriented proactivity can lead users to feel unappreciated, reducing their satisfaction and willingness to continue the conversation in human-computer interactions. To address this issue, we propose a User-oriented Proactive Chatbot (UPC) to enhance the user-oriented proactivity. Specifically, we first construct a critic to evaluate this proactivity inspired by the LLM-as-a-judge strategy. Given the scarcity of high-quality training data, we then employ the critic to guide dialogues between the chatbot and user agents, generating a corpus with enhanced user-oriented proactivity. To ensure the diversity of the user backgrounds, we introduce the ISCO-800, a diverse user background dataset for constructing user agents. Moreover, considering the communication difficulty varies among users, we propose an iterative curriculum learning method that trains the chatbot from easy-to-communicate users to more challenging ones, thereby gradually enhancing its performance. Experiments demonstrate that our proposed training method is applicable to different LLMs, improving user-oriented proactivity and attractiveness in open-domain dialogues. Code and appendix are available at github.com/wang678/LLM-UPC.

LLMs have demonstrated impressive performance across various language tasks. However, the strengths of LLMs can vary due to different architectures, model sizes, areas of training data, etc. Therefore, ensemble reasoning for the strengths of different LLM experts is critical to achieving consistent and satisfactory performance on diverse inputs across a wide range of tasks. However, existing LLM ensemble methods are either computationally intensive or incapable of leveraging complementary knowledge among LLM experts for various inputs. In this paper, we propose an efficient Dynamic Ensemble Reasoning paradigm, called DER to integrate the strengths of multiple LLM experts conditioned on dynamic inputs. Specifically, we model the LLM ensemble reasoning problem as a Markov Decision Process, wherein an agent sequentially takes inputs to request knowledge from an LLM candidate and passes the output to a subsequent LLM candidate. Moreover, we devise a reward function to train a DER-Agent to dynamically select an optimal answering route given the input questions, aiming to achieve the highest performance with as few computational resources as possible. Last, to fully transfer the expert knowledge from the prior LLMs, we develop a Knowledge Transfer Prompt that enables the subsequent LLM candidates to transfer complementary knowledge effectively. Experiments demonstrate that our method uses fewer computational resources to achieve better performance compared to state-of-the-art baselines. Code and appendix are available at https://github.com/Fhujinwu/DER.

Quantizing deep models prior to deployment is a widely adopted technique to speed up inference for various real-time applications, such as autonomous driving. However, quantized models often suffer from severe performance degradation in dynamic environments with potential domain shifts and this degradation is significantly more pronounced compared with their full-precision counterparts, as shown by our theoretical and empirical illustrations. To address the domain shift problem, test-time adaptation (TTA) has emerged as an effective solution by enabling models to learn adaptively from test data. Unfortunately, existing TTA methods are often impractical for quantized models as they typically rely on gradient backpropagation-an operation that is unsupported on quantized models due to vanishing gradients, as well as memory and latency constraints. In this paper, we focus on TTA for quantized models to improve their robustness and generalization ability efficiently. We propose a continual zeroth-order adaptation (ZOA) framework that enables efficient model adaptation using only two forward passes, eliminating the computational burden of existing methods. Moreover, we propose a domain knowledge management scheme to store and reuse different domain knowledge with negligible memory consumption, reducing the interference of different domain knowledge and fostering the knowledge accumulation during long-term adaptation. Experimental results on three classical architectures, including quantized transformer-based and CNN-based models, demonstrate the superiority of our methods for quantized model adaptation. On the quantized W6A6 ViT-B model, our ZOA is able to achieve a 5.0% improvement over the state-of-the-art FOA on ImageNet-C dataset. The source code is available at https://github.com/DengZeshuai/ZOA.

Qianyue Wang, Jinwu Hu, Zhengping Li, Yufeng Wang, Daiyuan Li, Yu Hu, Mingkui Tan. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Reinforcement Learning enables agents to learn optimal behaviors through interactions with environments. However, real-world environments are typically non-stationary, requiring agents to continuously adapt to new tasks and changing conditions. Although Continual Reinforcement Learning facilitates learning across multiple tasks, existing methods often suffer from catastrophic forgetting and inefficient knowledge utilization. To address these challenges, we propose Continual Knowledge Adaptation for Reinforcement Learning (CKA-RL), which enables the accumulation and effective utilization of historical knowledge. Specifically, we introduce a Continual Knowledge Adaptation strategy, which involves maintaining a task-specific knowledge vector pool and dynamically using historical knowledge to adapt the agent to new tasks. This process mitigates catastrophic forgetting and enables efficient knowledge transfer across tasks by preserving and adapting critical model parameters. Additionally, we propose an Adaptive Knowledge Merging mechanism that combines similar knowledge vectors to address scalability challenges, reducing memory requirements while ensuring the retention of essential knowledge. Experiments on three benchmarks demonstrate that the proposed CKA-RL outperforms state-of-the-art methods, achieving an improvement of 4.20% in overall performance and 8.02% in forward transfer. The source code is available at https://github.com/Fhujinwu/CKA-RL.

Drone Visual Active Tracking aims to autonomously follow a target object by controlling the motion system based on visual observations, providing a more practical solution for effective tracking in dynamic environments. However, accurate Drone Visual Active Tracking using reinforcement learning remains challenging due to the absence of a unified benchmark and the complexity of open-world environments with frequent interference. To address these issues, we pioneer a systematic solution. First, we propose DAT, the first open-world drone active air-to-ground tracking benchmark. It encompasses 24 city-scale scenes, featuring targets with human-like behaviors and high-fidelity dynamics simulation. DAT also provides a digital twin tool for unlimited scene generation. Additionally, we propose a novel reinforcement learning method called GC-VAT, which aims to improve the performance of drone tracking targets in complex scenarios. Specifically, we design a Goal-Centered Reward to provide precise feedback across viewpoints to the agent, enabling it to expand perception and movement range through unrestricted perspectives. Inspired by curriculum learning, we introduce a Curriculum-Based Training strategy that progressively enhances the tracking performance in complex environments. Besides, experiments on simulator and real-world images demonstrate the superior performance of GC-VAT, achieving a Tracking Success Rate of approximately 72% on the simulator. The benchmark and code are available at https://github.com/SHWplus/DAT_Benchmark.

AI-generated videos have achieved near-perfect visual realism (e.g., Sora), urgently necessitating reliable detection mechanisms. However, detecting such videos faces significant challenges in modeling high-dimensional spatiotemporal dynamics and identifying subtle anomalies that violate physical laws. In this paper, we propose a physics-driven AI-generated video detection paradigm based on probability flow conservation principles. Specifically, we propose a statistic called Normalized Spatiotemporal Gradient (NSG), which quantifies the ratio of spatial probability gradients to temporal density changes, explicitly capturing deviations from natural video dynamics. Leveraging pre-trained diffusion models, we develop an NSG estimator through spatial gradients approximation and motion-aware temporal modeling without complex motion decomposition while preserving physical constraints. Building on this, we propose an NSG-based video detection method (NSG-VD) that computes the Maximum Mean Discrepancy (MMD) between NSG features of the test and real videos as a detection metric. Last, we derive an upper bound of NSG feature distances between real and generated videos, proving that generated videos exhibit amplified discrepancies due to distributional shifts. Extensive experiments confirm that NSG-VD outperforms state-of-the-art baselines by 16.00\% in Recall and 10.75\% in F1-Score, validating the superior performance of NSG-VD. The source code is available at \url{https://github.com/ZSHsh98/NSG-VD}.

Test-time adaptation (TTA) seeks to tackle potential distribution shifts between training and testing data by adapting a given model w.r.t. any testing sample. This task is particularly important when the test environment changes frequently. Although some recent attempts have been made to handle this task, we still face two key challenges: 1) prior methods have to perform backpropagation for each test sample, resulting in unbearable optimization costs to many applications; 2) while existing TTA solutions can significantly improve the test performance on out-of-distribution data, they often suffer from severe performance degradation on in-distribution data after TTA (known as catastrophic forgetting). To this end, we have proposed an Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples for test-time entropy minimization. To alleviate forgetting, EATA introduces a Fisher regularizer estimated from test samples to constrain important model parameters from drastic changes. However, in EATA, the adopted entropy loss consistently assigns higher confidence to predictions even when the samples are underlying uncertain, leading to overconfident predictions that underestimate the data uncertainty. To tackle this, we further propose EATA with Calibration (EATA-C) to separately exploit the reducible model uncertainty and the inherent data uncertainty for calibrated TTA. Specifically, we compare the divergence between predictions from the full network and its sub-networks to measure the reducible model uncertainty, on which we propose a test-time uncertainty reduction strategy with divergence minimization loss to encourage consistent predictions instead of overconfident ones. To further re-calibrate predicting confidence on different samples, we utilize the disagreement among predicted labels as an indicator of the data uncertainty. Based on this, we devise a min-max entropy regularization to selectively increase and decrease predicting confidence for confidence re-calibration. Note that EATA-C and EATA are different on the adaptation objective, while EATA-C still benefits from the active sample selection criterion and anti-forgetting Fisher regularization proposed in EATA. Extensive experiments on image classification and semantic segmentation verify the effectiveness of our proposed methods.

Long video understanding has become a critical task in computer vision, driving advancements across numerous applications from surveillance to content retrieval. Existing video understanding methods suffer from two challenges when dealing with long video understanding: intricate long-context relationship modeling and interference from redundancy. To tackle these challenges, we introduce Fine-Detailed Video Story generation (FDVS), which interprets long videos into detailed textual representations. Specifically, to achieve fine-grained modeling of long-temporal content, we propose a Bottom-up Video Interpretation Mechanism that progressively interprets video content from clips to video. To avoid interference from redundant information in videos, we introduce a Semantic Redundancy Reduction mechanism that removes redundancy at both the visual and textual levels. Our method transforms long videos into hierarchical textual representations that contain multi-granularity information of the video. With these representations, FDVS is applicable to various tasks without any fine-tuning. We evaluate the proposed method across eight datasets spanning three tasks. The performance demonstrates the effectiveness and versatility of our method.

Visual grounding for remote sensing images (RSVG) is a fundamental vision-language task, which aims to locate the objects referred to by the natural language expression from the RS images. Natural language expressions often rely heavily on detailed directional information to describe target objects. Therefore, the thorough and effective utilization of spatial directional information is crucial for accurately locating the referred objects within complex RS images. However, most existing RSVG methods fail to fully leverage directional information, leading to suboptimal outcomes. This paper introduces a novel directional semantic enhanced method for RSVG, dubbed DSEVG. Specifically, we propose a scale-adaptive language-guided interaction (SLI) module that derives scale-specific language features through a hierarchical language adaptation mechanism. These scale-specific language features guide the visual backbone to extract visual features highly relevant to referring expressions. Furthermore, we present a directional semantic enhancement (DSE) module that implicitly enhances directional semantics in referring expressions by leveraging visual spatial information. It also employs an explicit spatial semantic alignment loss to supervise this process, generating localization prompts with enhanced directional representations. These prompts are injected into the queries of each decoder layer to guide the model in effectively utilizing directional semantics. Experimental results on the DIOR-RSVG and OPT-RSVG benchmark datasets validate the effectiveness of the proposed method and demonstrate state-of-the-art performance. Code is available at: https://github.com/WH231203/DSEVG.

Zero-shot skeleton-based action recognition aims to classify unseen skeleton-based human actions without prior exposure to such categories during training. This task is extremely challenging due to the difficulty in generalizing from known to unknown actions. Previous studies typically use two-stage training: pre-training skeleton encoders on seen action categories using cross-entropy loss and then aligning pre-extracted skeleton and text features, enabling knowledge transfer to unseen classes through skeleton-text alignment and language models' generalization. However, their efficacy is hindered by 1) insufficient discrimination for skeleton features, as the fixed skeleton encoder fails to capture necessary alignment information for effective skeleton-text alignment; 2) the neglect of alignment bias between skeleton and unseen text features during testing. To this end, we propose a prototype-guided feature alignment paradigm for zero-shot skeleton-based action recognition, termed PGFA. Specifically, we develop an end-to-end cross-modal contrastive training framework to improve skeleton-text alignment, ensuring sufficient discrimination for skeleton features. Additionally, we introduce a prototype-guided text feature alignment strategy to mitigate the adverse impact of the distribution discrepancy during testing. We provide a theoretical analysis to support our prototype-guided text feature alignment strategy and empirically evaluate our overall PGFA on three well-known datasets. Compared with the top competitor SMIE method, our PGFA achieves absolute accuracy improvements of 22.96%, 12.53%, and 18.54% on the NTU-60, NTU-120, and PKU-MMD datasets, respectively.

Vision-and-Language Navigation (VLN) requires an agent to follow given instructions to navigate. Despite the significant progress, the model trained on seen environments has a performance drop on unseen environments due to distribution shift. To improve the generalization, existing method attempts to apply test-time adaptation to VLN. However, it needs to access the training data and all testing data for updating the model before inference. The setting is not suitable for the real application because it is hard for the agent to access training data and all testing data when the agent is applied in a new environment. In this paper, we consider a more practical setting with source-free and online-inference test-time adaption. In other words, the model can only access one testing sample for test-time adaptation. In this setting, the model may suffer from catastrophic forgetting of the learned knowledge and unstable parameter update issues. To solve these challenges, we propose an elastic adaptation model (EAM) that consists of an auxiliary decision model and a sample replay mechanism. We use the online testing samples to adapt the auxiliary decision model to new environments, which cooperates with the frozen original model to make better action decisions. The sample replay mechanism stores the historical testing samples to make the adaptation process more stable. Our method is model-agnostic and is effortless to be applied to most existing methods. Experimental results show that our method achieves stable performance improvement based on three existing methods on three VLN benchmark datasets.

Test-time adaptation (TTA) is a task that continually adapts a pre-trained source model to the target domain during inference. One popular approach involves fine-tuning model with cross-entropy loss according to estimated pseudo-labels. However, its performance is significantly affected by noisy pseudo-labels. This study reveals that minimizing the classification error of each sample causes the cross-entropy loss's vulnerability to label noise. To address this issue, we propose a novel Decoupled Prototype Learning (DPL) method that features prototype-centric loss computation. First, we decouple the optimization of class prototypes. For each class prototype, we reduce its distance with positive samples and enlarge its distance with negative samples in a contrastive manner. This strategy prevents the model from overfitting to noisy pseudo-labels. Second, we propose a memory-based strategy to enhance DPL's robustness for the small batch sizes often encountered in TTA. We update each class's pseudo-feature from a memory in a momentum manner and insert an additional DPL loss. Finally, we introduce a consistency regularization-based approach to leverage samples with unconfident pseudo-labels. This approach transfers feature styles of samples with unconfident pseudo-labels to those with confident pseudo-labels. Thus, more reliable samples for TTA are created. The experimental results demonstrate that our methods achieve state-of-the-art performance on domain generalization benchmarks, and reliably improve the performance of self-training-based methods on image corruption benchmarks.

Visual Question Answering (VQA) is a prevalent task that can facilitate the perception of the real world by the visually impaired. However, many VQA models tend to rely on superficial correlations in datasets for predictions rather than genuine reasoning, limiting their real-world applicability. While existing methods address this issue by incorporating debiasing strategies during training, they typically assume prior knowledge of out-of-distribution (OOD) test sets and then tailor debiasing strategies and select optimal models on the basis of the OOD samples. This reliance on OOD test data, however, is unrealistic in practical applications. To address this, some works introduce test-time adaptation techniques to mitigate dataset shifts during model deployment. Despite their potential, these methods risk catastrophic forgetting as they update models at test time without access to the ground-truth answers or the source data. An emerging solution involves leveraging the extensive knowledge embedded in Large Language Models (LLMs) to support reasoning tasks, yet their language-only input restricts flexibility in multimodal tasks. To bridge this gap, we propose leveraging the zero-shot capability of Multimodal Large Language Models (MLLMs). To optimise computational efficiency, we introduce a novel VQA Collaborative Inference framework (VQA-CI) that integrates MLLMs (e.g.BLIP-2 Flan T5) with VQA specialists (e.g.UpDn). This framework initially processes samples through VQA specialists and subsequently determines the necessity for re-evaluation with MLLMs based on predefined bias and reliability indicators. Experiments on the GQA-OOD and VQA-CP v2 datasets show that our VQA-CI achieves significant performance gains, with accuracy improvements of around 6% over state-of-the-art methods, underscoring the effectiveness of our VQA-CI.

In multiagent systems, learning optimal behavior policies for individual agents remains a challenging yet crucial task. While recent research has made strides in this area, the issue of when agents should maintain consistent behaviors with one another is still not adequately addressed. This article proposes a novel approach to enable agents to autonomously decide whether their behaviors should align with those of their peers by leveraging intrinsic rewards to optimize their policies. We define behavior consistency as the divergence between the actions taken by two agents given the same observations. To encourage agents to be aware of each other's behaviors, we propose dynamic consistency-based intrinsic reward (DCIR), which guides agents in determining when to synchronize their behaviors. In addition, we introduce a dynamic scaling network (DSN) that provides learnable scaling factors at each time step, enabling agents to dynamically decide the extent of rewarding consistent behavior. Our method is evaluated on environments including Multiagent Particle, Google Research Football, and StarCraft II Micromanagement. Experimental results demonstrate its effectiveness in learning optimal policies.

The task of financial analysis primarily encompasses two key areas: stock trend prediction and the corresponding financial question answering. Currently, machine learning and deep learning algorithms (ML&DL) have been widely applied for stock trend predictions, leading to significant progress. However, these methods fail to provide reasons for predictions, lacking interpretability and reasoning processes. Also, they can not integrate textual information such as financial news or reports. Meanwhile, large language models (LLM) have remarkable textual understanding and generation ability. But due to the scarcity of financial training datasets and limited integration with real-time knowledge, LLM still suffer from hallucinations and unable to keep up with the latest information. To tackle these challenges, we first release AlphaFin datasets, combining traditional research datasets, real-time financial data, and handwritten chain-of-thought (CoT) data. It has positive impact on training LLM for completing financial analysis. We then use AlphaFin datasets to benchmark a state-of-the-art method, called Stock-Chain, for effectively tackling the financial analysis task, which integrates retrieval-augmented generation (RAG) techniques. Extensive experiments are conducted to demonstrate the effectiveness of our framework on financial analysis.

Novel view synthesis aims to generate new view images of a given view image collection. Recent attempts address this problem relying on 3D geometry priors (e.g., shapes, sizes, and positions) learned from multi-view images. However, such methods encounter the following limitations: 1) they require a set of multi-view images as training data for a specific scene (e.g., face, car or chair), which is often un-available in many real-world scenarios; 2) they fail to ex-tract the geometry priors from single-view images due to the lack of multi-view supervision. In this paper, we pro-pose a Geometry-enhanced NeRF (G-NeRF), which seeks to enhance the geometry priors by a geometry-guided multi-view synthesis approach, followed by a depth-aware training. In the synthesis process, inspired that existing 3D GAN models can unconditionally synthesize high-fidelity multi-view images, we seek to adopt off-the-shelf 3D GAN models, such as EG3D, as a free source to provide geometry priors through synthesizing multi-view data. Simultaneously, to further improve the geometry quality of the synthetic data, we introduce a truncation method to effectively sample la-tent codes within 3D GAN models. To tackle the absence of multi-view supervision for single-view images, we design the depth-aware training approach, incorporating a depth-aware discriminator to guide geometry priors through depth maps. Experiments demonstrate the effectiveness of our method in terms of both qualitative and quantitative results.

Reconstructing 3D clothed human involves creating a detailed geometry of individuals in clothing, with applications ranging from virtual try-on, movies, to games. To enable practical and widespread applications, recent ad-vances propose to generate a clothed human from an RGB image. However, they struggle to reconstruct detailed and robust avatars simultaneously. We empirically find that the high-frequency (HF) and low-frequency (LF) information from a parametric model has the potential to enhance geom-etry details and improve robustness to noise, respectively. Based on this, we propose HiLo, namely clothed human reconstruction with high- and low-frequency information, which contains two components. 1) To recover detailed ge-ometry using HF information, we propose a progressive HF Signed Distance Function to enhance the detailed 3D ge-ometry of a clothed human. We analyze that our progres-sive learning manner alleviates large gradients that hin-der model convergence. 2) To achieve robust reconstruction against inaccurate estimation of the parametric model by using LF information, we propose a spatial interaction implicit function. This function effectively exploits the com-plementary spatial information from a low-resolution voxel grid of the parametric model. Experimental results demon-strate that HiLo outperforms the state-of-the-art methods by 10.43% and 9.54% in terms of Chamfer distance on the Thuman2.0 and CAPE datasets, respectively. Additionally, HiLo demonstrates robustness to noise from the parametric model, challenging poses, and various clothing styles.11Code link https://github.com/YifYang993/HiLo.git

The conventional deep learning paradigm often involves training a deep model on a server and then deploying the model or its distilled ones to resource-limited edge devices. Usually, the models shall remain fixed once deployed (at least for some period) due to the potential high cost of model adaptation for both the server and edge sides. However, in many real-world scenarios, the test environments may change dynamically (known as distribution shifts), which often results in degraded performance. Thus, one has to adapt the edge models promptly to attain promising performance. Moreover, with the increasing data collected at the edge, this paradigm also fails to further adapt the cloud model for better performance. To address these, we encounter two primary challenges: 1) the edge model has limited computation power and may only support forward propagation; 2) the data transmission budget between cloud and edge devices is limited in latency-sensitive scenarios. In this paper, we establish a Cloud-Edge Elastic Model Adaptation (CEMA) paradigm in which the edge models only need to perform forward propagation and the edge models can be adapted online. In our CEMA, to reduce the communication burden, we devise two criteria to exclude unnecessary samples from uploading to the cloud, i.e., dynamic unreliable and low-informative sample exclusion. Based on the uploaded samples, we update and distribute the affine parameters of normalization layers by distilling from the stronger foundation model to the edge model with a sample replay strategy. Extensive experimental results on ImageNet-C and ImageNet-R verify the effectiveness of our CEMA.

Large language models (LLMs) such as ChatGPT have exhibited remarkable performance in generating human-like texts. However, machine-generated texts (MGTs) may carry critical risks, such as plagiarism issues and hallucination information. Therefore, it is very urgent and important to detect MGTs in many situations. Unfortunately, it is challenging to distinguish MGTs and human-written texts because the distributional discrepancy between them is often very subtle due to the remarkable performance of LLMS. In this paper, we seek to exploit \textit{maximum mean discrepancy} (MMD) to address this issue in the sense that MMD can well identify distributional discrepancies. However, directly training a detector with MMD using diverse MGTs will incur a significantly increased variance of MMD since MGTs may contain \textit{multiple text populations} due to various LLMs. This will severely impair MMD's ability to measure the difference between two samples. To tackle this, we propose a novel \textit{multi-population} aware optimization method for MMD called MMD-MP, which can \textit{avoid variance increases} and thus improve the stability to measure the distributional discrepancy. Relying on MMD-MP, we develop two methods for paragraph-based and sentence-based detection, respectively. Extensive experiments on various LLMs, \eg, GPT2 and ChatGPT, show superior detection performance of our MMD-MP.

In this paper, we propose test-time Collaborative Lifelong Adaptation (CoLA), which is a general paradigm that can be incorporated with existing advanced TTA methods to boost the adaptation performance and efficiency in a multi-device collaborative manner. Specifically, we maintain and store a set of device-shared _domain knowledge vectors_, which accumulates the knowledge learned from all devices during their lifelong adaptation process. Based on this, CoLA conducts two collaboration strategies for devices with different computational resources and latency demands. 1) Knowledge reprogramming learning strategy jointly learns new domain-specific model parameters and a reweighting term to reprogram existing shared domain knowledge vectors, termed adaptation on _principal agents_. 2) Similarity-based knowledge aggregation strategy solely aggregates the knowledge stored in shared domain vectors according to domain similarities in an optimization-free manner, termed adaptation on _follower agents_. Experiments verify that CoLA is simple but effective, which boosts the efficiency of TTA and demonstrates remarkable superiority in collaborative, lifelong, and single-domain TTA scenarios, e.g., on follower agents, we enhance accuracy by over 30\% on ImageNet-C while maintaining nearly the same efficiency as standard inference. The source code is available at https://github.com/Cascol-Chen/COLA.

Image captioning is one of the fundamental problems of computer vision and has drawn great attention over the years. However, most existing methods in image captioning focus on improving the quality of the image captions, while ignoring the ability of controlling the caption style. In this work, we aim to improve the controllability of image captioning methods, especially, by choosing to describe the image either roughly or in detail. We find this can be achieved by adding a simple length level embedding into existing models, which enables them to generate length-controllable captions describing the image at a specified level of detail, and further improve the diversity. On top of it, we propose to learn a length-level reranking transformer that captures the correlation between the semantic complexities of the image and text modalities, which can be used to select the most suitable length level for each image to make the captions informative while not being redundant. Moreover, when the length of the generated captions grows, existing methods usually suffer from a linearly increased computational complexity due to their autoregressive (AR) nature. To aid this, we devise a non-autoregressive (NAR) approach that generates captions in a length-irrelevant complexity. Besides, we propose a refinement-enhanced sequence training scheme and also adopt a sequence-level knowledge distillation technique for the training of our NAR model to bridge its performance gap with the AR models. In the experiments, our length-controllable models not only achieve SOTA performance in terms of caption quality on the MS COCO dataset but more importantly, generate controllable and diverse image captions. Specifically, our NAR model outperforms the AR baselines in terms of controllability and diversity, and also significantly improves the decoding efficiency for longer captions. By further applying the sequence-level training schemes, the caption quality of our NAR model improves clearly and is competitive with the state-of-the-art AR baselines.

Deep neural networks have exhibited remarkable performance in image super-resolution (SR) tasks by learning a mapping from low-resolution (LR) images to high-resolution (HR) images. However, the SR problem is typically an ill-posed problem and existing methods would come with several limitations. First, the possible mapping space of SR can be extremely large since there may exist many different HR images that can be super-resolved from the same LR image. As a result, it is hard to directly learn a promising SR mapping from such a large space. Second, it is often inevitable to develop very large models with extremely high computational cost to yield promising SR performance. In practice, one can use model compression techniques to obtain compact models by reducing model redundancy. Nevertheless, it is hard for existing model compression methods to accurately identify the redundant components due to the extremely large SR mapping space. To alleviate the first challenge, we propose a dual regression learning scheme to reduce the space of possible SR mappings. Specifically, in addition to the mapping from LR to HR images, we learn an additional dual regression mapping to estimate the downsampling kernel and reconstruct LR images. In this way, the dual mapping acts as a constraint to reduce the space of possible mappings. To address the second challenge, we propose a dual regression compression (DRC) method to reduce model redundancy in both layer-level and channel-level based on channel pruning. Specifically, we first develop a channel number search method that minimizes the dual regression loss to determine the redundancy of each layer. Given the searched channel numbers, we further exploit the dual regression manner to evaluate the importance of channels and prune the redundant ones. Extensive experiments show the effectiveness of our method in obtaining accurate and efficient SR models.

We study multi-sensor fusion for 3D semantic segmentation that is important to scene understanding for many applications, such as autonomous driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to effectively exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinate using perspective projection. In this way, we can process both inputs from LiDAR and cameras in 2D space while preventing the information loss of RGB images. Then, we propose a two-stream network that consists of a LiDAR stream and a camera stream to extract features from the two modalities, separately. The extracted features are fused by effective residual-based fusion modules. Moreover, we introduce additional perception-aware losses to measure the perceptual difference between the two modalities. Last, we propose an improved version of PMF,i.e., EPMF, which is more efficient and effective by optimizing data pre-processing and network architecture under perspective projection. Specifically, we propose cross-modal alignment and cropping to obtain tight inputs and reduce unnecessary computational costs. We then explore more efficient contextual modules under perspective projection and fuse the LiDAR features into the camera stream to boost the performance of the two-stream network. Extensive experiments on benchmark data sets show the superiority of our method. For example, on nuScenes test set, our EPMF outperforms the state-of-the-art method,i.e., RangeFormer, by0.9%in mIoU. Compared to PMF, EPMF also achieves2.06× acceleration with2.0%improvement in mIoU. Our source code is available athttps://github.com/ICEORY/PMF.

Neural Architecture Search (NAS) aims to automatically find effective architectures within a predefined search space. However, the search space is often extremely large. As a result, directly searching in such a large search space is non-trivial and also very time-consuming. To address the above issues, in each search step, we seek to limit the search space to a small but effective subspace to boost both the search performance and search efficiency. To this end, we propose a novel Neural Architecture Search method via Dominative Subspace Mining (DSM-NAS) that finds promising architectures in automatically mined subspaces. Specifically, we first perform a global search,i.e., dominative subspace mining, to find a good subspace from a set of candidates. Then, we perform a local search within the mined subspace to find effective architectures. More critically, we further boost search performance by taking well-designed/ searched architectures to initialize candidate subspaces. Experimental results demonstrate that DSM-NAS not only reduces the search cost but also discovers better architectures than state-of-the-art methods in various benchmark search spaces.

Network pruning has been widely studied to reduce the complexity of deep neural networks (DNNs) and hence speed up their inference. Unfortunately, most existing pruning methods ignore the changes in the model’s robustness before and after pruning, which makes pruned models vulnerable under dynamically perturbed environments (e.g., autonomous driving). Only a few works have explored the robustness of pruned models against adversarial attacks that significantly differ from perturbations in real-world scenarios. To bridge the gap between real-world applications and existing studies, in this work, we propose an adversarial pruning scheme, which automatically identifies and preserves robust channels to obtain robust pruned models that are suitable for practical deployment in dynamically perturbed environments. Specifically, to simulate real-world perturbations, we first employ multi-type adversarial attack samples and adversarial perturbation samples generated by an adversarial perturbation generator to create mixed noise samples. Then, we propose a plug-and-play feature scoring module and a novel contribution difference loss to evaluate the robustness of intermediate features dynamically. Next, to leverage robust intermediate features to identify robust channels, we have developed a simple but effective gating mechanism that evaluates the robustness of channels and preserves robust channels during training. Lastly, we compress the model in a layer-wise or block-wise manner. Compared to existing methods, our scheme enhances the robustness of the pruned model in a broader sense, making it better able to against dynamic perturbations in the real world. Extensive experimental results on well-known dataset benchmarks and popular network architectures demonstrate the effectiveness of our method.

Model quantization is a prevalent method to compress and accelerate neural networks. Most existing quantization methods usually require access to real data to improve the performance of quantized models, which is often infeasible in some scenarios with privacy and security concerns. Recently, data-free quantization has been widely studied to solve the challenge of not having access to real data by generating synthetic data, among which generator-based data-free quantization is an important type. Previous generator-based methods focus on improving the performance of quantized models by optimizing the spatial distribution of synthetic data, while ignoring the study of changes in synthetic data from a temporal perspective. In this work, we reveal that generator-based data-free quantization methods usually suffer from the issue that synthetic data show homogeneity in the mid-to-late stages of the generation process due to the stagnation of the generator update, which hinders further improvement of the performance of the quantization model. To solve the above issue, we propose introducing the discrepancy between the full-precision and quantized models as new supervision information to update the generator. Specifically, we propose a simple yet effective adversarial Gaussian-margin loss, which promotes continuous updating of the generator by adding more supervision information to the generator when the discrepancy between the full-precision and quantized models is small, thereby generating heterogeneous synthetic data. Moreover, to mitigate the homogeneity of the synthetic data further, we augment the synthetic data with linear interpolation. Our proposed method can also promote the performance of other generator-based data-free quantization methods. Extensive experimental results show that our proposed method achieves superior performances for various settings on data-free quantization, especially in ultra-low-bit settings, such as 3-bit.

In this paper, a novel LiDAR-Camera Alignment (LCA) method using homogeneous local-global spatial aware representation is proposed. Compared with the state-of-the-art methods (e.g., LCCNet), our proposition holds 2 main superiorities. First, homogeneous multi-modality representation learned with a uniform CNN model is applied along the iterative prediction stages, instead of the state-of-the-art heterogeneous counterparts extracted from the separated modality-wise CNN models within each stage. In this way, the model size can be significantly decreased (e.g., 12.39M (ours) vs. 333.75M (LCCNet)). Meanwhile, within our proposition the interaction between LiDAR and camera data is built during feature learning to better exploit the descriptive clues, which has not been well concerned by the existing approaches. Secondly, we propose to equip the learned LCA representation with local-global spatial aware capacity via encoding CNN’s local convolutional features with Transformer’s non-local self-attention manner. Accordingly, the local fine details and global spatial context can be jointly captured by the encoded local features. And, they will be jointly used for LCA. On the other hand, the existing methods generally choose to reveal the global spatial property via intuitively concatenating the local features. Additionally at the initial LCA stage, LiDAR is roughly aligned with camera by our pre-alignment method, according to the point distribution characteristics of its 2D projection version with the initial extrinsic parameters. Although its structure is simple, it can essentially alleviate LCA’s difficulty for the consequent stages. To better optimize LCA, a novel loss function that builds the correlation between translation and rotation loss items is also proposed. The experiments on KITTI data verifies the superiority of our proposition both on effectiveness and efficiency. The source code will be released athttps://github.com/Zaf233/Light-weight-LCAupon acceptance.

Many of the tissues/lesions in the medical images may be ambiguous. Therefore, medical segmentation is typically annotated by a group of clinical experts to mitigate personal bias. A common solution to fuse different annotations is the majority vote, e.g., taking the average of multiple labels. However, such a strategy ignores the difference between the grader expertness. Inspired by the observation that medical image segmentation is usually used to assist the disease diagnosis in clinical practice, we propose the diagnosis-first principle, which is to take disease diagnosis as the criterion to calibrate the inter-observer segmentation uncertainty. Following this idea, a framework named Diagnosis-First segmentation Framework (DiFF) is proposed. Specifically, DiFF will first learn to fuse the multi-rater segmentation labels to a single ground-truth which could maximize the disease diagnosis performance. We dubbed the fused ground-truth as Diagnosis-First Ground-truth (DF-GT). Then, the Take and Give Model (T&G Model) to segment DF-GT from the raw image is proposed. With the T&G Model, DiFF can learn the segmentation with the calibrated uncertainty that facilitate the disease diagnosis. We verify the effectiveness of DiFF on three different medical segmentation tasks: optic-disc/optic-cup (OD/OC) segmentation on fundus images, thyroid nodule segmentation on ultrasound images, and skin lesion segmentation on dermoscopic images. Experimental results show that the proposed DiFF can effectively calibrate the segmentation uncertainty, and thus significantly facilitate the corresponding disease diagnosis, which outperforms previous state-of-the-art multi-rater learning methods.

Visual question answering (VQA) is a prevalent task in real-world, and plays an essential role in helping the blind understand the physical world. However, due to the real-world complexity, VQA test samples may come from a different distribution from the training data, resulting in unavoidable performance degradation. This similar issue also exists in the image recognition field, in which one most recent effective solutions is a test-time adaptation (TTA). TTA adapts a trained model at test time using only test samples, which provides a new idea to alleviate the analogous issue in VQA. However, naively introducing existing TTA methods (e.g.,test-time entropy minimisation) into VQA is imperfect and achieves only marginal performance gain. The reason is that prior methods do not consider the special nature of the VQA problem and ignore that 1) the biased samples in the dataset may have negative effects on test-time model adaptation, and 2) the model may have captured the biases in the dataset. In this paper, we propose Test-time Debiased Self-supervised (TDS) learning objectives for VQA model adaptation. Specifically, we minimise the entropy for those unbiased test samples. To identify these samples, we construct a negative sample for each test sample, and regard the test samples as unbiased if the output answers are different when feeding the test sample and the counterpart negative sample into the VQA model. Meanwhile, we also remove those samples with high prediction entropy from adaptation, making the test-time gradients more reliable. To hinder the model from excessively fitting the superficial correlations of the biased sample, we adopt the biased samples and the counterpart negative samples to assist the adaptation. Extensive experiments on the VQA-CP v1 and VQA-CP v2 datasets demonstrate the effectiveness of our TDS.

Text-to-image synthesis aims to generate realistic images conditioned on text descriptions. To fuse text information into synthesized images, conditional affine transformations (CATs), such as conditional batch normalization (CBN) and conditional instance normalization (CIN), are usually used to predict batch statistics of different layers. However, ordinary CAT blocks control the batch statistics independently disregarding the consistency among neighboring layers. To address the above issue, we propose a new fusion approach names recurrent affine transformation (RAT) for synthesizing images conditioned on text information. RAT connects all the CAT blocks with recurrent connections for explicitly fitting the temporal consistency between CAT blocks. To verify the effectiveness of RAT, we propose a novel visualization method to show how generative adversarial network (GAN) fuses conditional information. Our microscopic and macroscopic visualizations not only demonstrate the effectiveness of RAT but also turn out to be a useful perspective to analyze how GAN fuses conditional information. In addition, we propose a more stable spatial attention mechanism for the discriminator, which helps the text description to supervise the generator to synthesize more relevant image contents. Extensive experiments on the CUB, Oxford-102, and COCO datasets demonstrate the proposed model's superiority in comparison to state-of-the-art models. Our code is available onhttps://github.com/senmaoy/RAT-GAN.

Visual Relational Reasoning is the basis of many vision-and-language based tasks (e.g., visual question answering and referring expression comprehension). In this article, we regard the complex referring expression comprehension (c-REF) task as the reasoning basis, in which c-REF seeks to localise a target object in an image guided by a complex query. Such queries often contain complex logic and thus impose two critical challenges for reasoning: (i) Comprehending the complex queries is difficult since these queries usually refer to multiple objects and their relationships; (ii) Reasoning among multiple objects guided by the queries and then localising the target correctly are non-trivial. To address the above challenges, we propose a Transformer-based Relational Inference Network (Trans-RINet). Specifically, to comprehend the queries, we mimic the language-comprehending mechanism of humans, and devise a language decomposition module to decompose the queries into four types, i.e., basic attributes, absolute location, visual relationship and relative location. We further devise four modules to address the corresponding information. In each module, we consider the intra-(i.e., between the objects) and inter-modality relationships(i.e., between the queries and objects) to improve the reasoning ability. Moreover, we construct a relational graph to represent the objects and their relationships, and devise a multi-step reasoning method to progressively understand the complex logic. Since each type of the queries is closely related, we let each module interact with each other before making a decision. Extensive experiments on the CLEVR-Ref+, Ref-Reasoning, and CLEVR-CoGenT datasets demonstrate the superior reasoning performance of our Trans-RINet.

Zero-shot quantization (ZSQ) is promising for compressing and accelerating deep neural networks when the data for training full-precision models are inaccessible. In ZSQ, network quantization is performed using synthetic samples, thus, the performance of quantized models depends heavily on the quality of synthetic samples. Nonetheless, we find that the synthetic samples constructed in existing ZSQ methods can be easily fitted by models. Accordingly, quantized models obtained by these methods suffer from significant performance degradation on hard samples. To address this issue, we propose HArd sample Synthesizing and Training (HAST). Specifically, HAST pays more attention to hard samples when synthesizing samples and makes synthetic samples hard to fit when training quantized models. HAST aligns features extracted by full-precision and quantized models to ensure the similarity between features extracted by these two models. Extensive experiments show that HAST significantly outperforms existing ZSQ methods, achieving performance comparable to models that are quantized with real data.

How to learn discriminative video representation from unlabeled videos is challenging but crucial for video analysis. The latest attempts seek to learn a representation model by predicting the appearance contents in the masked regions. However, simply masking and recovering appearance contents may not be sufficient to model temporal clues as the appearance contents can be easily reconstructed from a single frame. To overcome this limitation, we present Masked Motion Encoding (MME), a new pretraining paradigm that reconstructs both appearance and motion information to explore temporal clues. In MME, we focus on addressing two critical challenges to improve the representation performance: 1) how to well represent the possible long-term motion across multiple frames; and 2) how to obtain fine-grained temporal clues from sparsely sampled videos. Motivated by the fact that human is able to recognize an action by tracking objects' position changes and shape changes, we propose to reconstruct a motion trajectory that represents these two kinds of change in the masked regions. Besides, given the sparse video input, we enforce the model to reconstruct dense motion trajectories in both spatial and temporal dimensions. Pre-trained with our MME paradigm, the model is able to anticipate long-term and fine-grained motion details. Code is available at https://github.com/XinyuSun/MME.

Vision-and-language navigation (VLN) requires an embodied agent to navigate in realistic 3D environments using natural language instructions. Existing VLN methods suffer from training on small-scale environments or unreasonable path-instruction datasets, limiting the generalization to unseen environments. There are massive house tour videos on YouTube, providing abundant real navigation experiences and layout information. However, these videos have not been explored for VLN before. In this paper, we propose to learn an agent from these videos by creating a large-scale dataset which comprises reasonable path-instruction pairs from house tour videos and pre-training the agent on it. To achieve this, we have to tackle the challenges of automatically constructing path-instruction pairs and exploiting real layout knowledge from raw and unlabeled videos. To address these, we first leverage an entropy-based method to construct the nodes of a path trajectory. Then, we propose an action-aware generator for generating instructions from unlabeled trajectories. Last, we devise a trajectory judgment pretext task to encourage the agent to mine the layout knowledge. Experimental results show that our method achieves state-of-the-art performance on two popular benchmarks (R2R and REVERIE). Code is available at https://github.com/JeremyLinky/YouTube-VLN

We study the task of weakly-supervised point cloud semantic segmentation with sparse annotations (e.g., less than 0.1% points are labeled), aiming to reduce the expensive cost of dense annotations. Unfortunately, with extremely sparse annotated points, it is very difficult to extract both contextual and object information for scene understanding such as semantic segmentation. Motivated by masked modeling (e.g., MAE) in image and video representation learning, we seek to endow the power of masked modeling to learn contextual information from sparsely-annotated points. However, directly applying MAE to 3D point clouds with sparse annotations may fail to work. First, it is nontrivial to effectively mask out the informative visual context from 3D point clouds. Second, how to fully exploit the sparse annotations for context modeling remains an open question. In this paper, we propose a simple yet effective Contextual Point Cloud Modeling (CPCM) method that consists of two parts: a region-wise masking (Region-Mask) strategy and a contextual masked training (CMT) method. Specifically, RegionMask masks the point cloud continuously in geometric space to construct a meaningful masked prediction task for subsequent context learning. CMT disentangles the learning of supervised segmentation and unsupervised masked context prediction for effectively learning the very limited labeled points and mass unlabeled points, respectively. Extensive experiments on the widely-tested ScanNet V2 and S3DIS benchmarks demonstrate the superiority of CPCM over the state-of-the-art.

Neural Radiance Fields (NeRF) is a revolutionary approach for rendering scenes by sampling a single ray per pixel and it has demonstrated impressive capabilities in novel-view synthesis from static scene images. However, in practice, we usually need to recover NeRF from unconstrained image collections, which poses two challenges: 1) the images often have dynamic changes in appearance because of different capturing time and camera settings; 2) the images may contain transient objects such as humans and cars, leading to occlusion and ghosting artifacts. Conventional approaches seek to address these challenges by locally utilizing a single ray to synthesize a color of a pixel. In contrast, humans typically perceive appearance and objects by globally utilizing information across multiple pixels. To mimic the perception process of humans, in this paper, we propose Cross-Ray NeRF (CR-NeRF) that leverages interactive information across multiple rays to synthesize occlusion-free novel views with the same appearances as the images. Specifically, to model varying appearances, we first propose to represent multiple rays with a novel cross-ray feature and then recover the appearance by fusing global statistics, i.e., feature covariance of the rays and the image appearance. Moreover, to avoid occlusion introduced by transient objects, we propose a transient objects handler and introduce a grid sampling strategy for masking out the transient objects. We theoretically find that leveraging correlation across multiple rays promotes capturing more global information. Moreover, extensive experimental results on large real-world datasets verify the effectiveness of CR-NeRF. The code and data can be found at https://github.com/YifYang993/CR-NeRF-PyTorch.git.

Heterogeneous Information Network(HIN) collective classification aims to classify one type of node, which is associated with multiple types of nodes through multiple types of relations. Previous studies have revealed that exploiting the relative importance of relation types is quite useful for improving node classification performance. We propose a Tensor-based Markov chain (T-Mark) model to improve the nodes classification accuracy by predicting the labels for unlabeled nodes and the importance ranking of relationship types automatically and simultaneously. Specifically, we build two tensor equations according to the HIN structure and content similarities among nodes of both labeled and unlabeled data. Consequently, We solve the semi-supervised T-Mark model by using an iterative process until obtaining two stationary distributions for labels and relation types. Experimental results on several real-world datasets demonstrate the effectiveness of T-Mark.

Portfolio Selection is an important real-world financial task and has attracted extensive attention in artificial intelligence communities. This task, however, has two main difficulties: (i) the non-stationary price series and complex asset correlations make the learning of feature representation very hard; (ii) the practicality principle in financial markets requires controlling both transaction and risk costs. Most existing methods adopt handcraft features and/or consider no constraints for the costs, which may make them perform unsatisfactorily and fail to control both costs in practice. In this paper, we propose a cost-sensitive portfolio selection method with deep reinforcement learning. Specifically, a novel two-stream portfolio policy network is devised to extract both price series patterns and asset correlations, while a new cost-sensitive reward function is developed to maximize the accumulated return and constrain both costs via reinforcement learning. We theoretically analyze the near-optimality of the proposed reward, which shows that the growth rate of the policy regarding this reward function can approach the theoretical optimum. We also empirically evaluate the proposed method on real-world datasets. Promising results demonstrate the effectiveness and superiority of the proposed method in terms of profitability, cost-sensitivity and representation abilities.

Adversarial detection aims to determine whether a given sample is an adversarial one based on the discrepancy between natural and adversarial distributions. Unfortunately, estimating or comparing two data distributions is extremely difficult, especially in high-dimension spaces. Recently, the gradient of log probability density (a.k.a., score) w.r.t. the sample is used as an alternative statistic to compute. However, we find that the score is sensitive in identifying adversarial samples due to insufficient information with one sample only. In this paper, we propose a new statistic called expected perturbation score (EPS), which is essentially the expected score of a sample after various perturbations. Specifically, to obtain adequate information regarding one sample, we perturb it by adding various noises to capture its multi-view observations. We theoretically prove that EPS is a proper statistic to compute the discrepancy between two samples under mild conditions. In practice, we can use a pre-trained diffusion model to estimate EPS for each sample. Last, we pro- pose an EPS-based adversarial detection (EPS- AD) method, in which we develop EPS-based maximum mean discrepancy (MMD) as a metric to measure the discrepancy between the test sample and natural samples. We also prove that the EPS-based MMD between natural and adversarial samples is larger than that among natural samples. Extensive experiments show the superior adversarial detection performance of our EPS-AD.

Image super-resolution (SR) aims to learn a mapping from low-resolution (LR) to high-resolution (HR) using paired HR-LR training images. Conventional SR methods typically gather the paired training data by synthesizing LR images from HR images using a predetermined degradation model, e.g., Bicubic down-sampling. However, the realistic degradation type of test images may mismatch with the training-time degradation type due to the dynamic changes of the real-world scenarios, resulting in inferior-quality SR images. To address this, existing methods attempt to estimate the degradation model and train an image-specific model, which, however, is quite time-consuming and impracticable to handle rapidly changing domain shifts. Moreover, these methods largely concentrate on the estimation of one degradation type (e.g., blur degradation), overlooking other degradation types like noise and JPEG in real-world test-time scenarios, thus limiting their practicality. To tackle these problems, we present an efficient test-time adaptation framework for SR, named SRTTA, which is able to quickly adapt SR models to test domains with different/unknown degradation types. Specifically, we design a second-order degradation scheme to construct paired data based on the degradation type of the test image, which is predicted by a pre-trained degradation classifier. Then, we adapt the SR model by implementing feature-level reconstruction learning from the initial test image to its second-order degraded counterparts, which helps the SR model generate plausible HR images. Extensive experiments are conducted on newly synthesized corrupted DIV2K datasets with 8 different degradations and several real-world datasets, demonstrating that our SRTTA framework achieves an impressive improvement over existing methods with satisfying speed. The source code is available at https://github.com/DengZeshuai/SRTTA.

Learning to navigate to an image-specified goal is an important but challenging task for autonomous systems like household robots. The agent is required to well understand and reason the location of the navigation goal from a picture shot in the goal position. Existing methods try to solve this problem by learning a navigation policy, which captures semantic features of the goal image and observation image independently and lastly fuses them for predicting a sequence of navigation actions. However, these methods suffer from two major limitations. 1) They may miss detailed information in the goal image, and thus fail to reason the goal location. 2) More critically, it is hard to focus on the goal-relevant regions in the observation image, because they attempt to understand observation without goal conditioning. In this paper, we aim to overcome these limitations by designing a Fine-grained Goal Prompting (\sexyname) method for image-goal navigation. In particular, we leverage fine-grained and high-resolution feature maps in the goal image as prompts to perform conditioned embedding, which preserves detailed information in the goal image and guides the observation encoder to pay attention to goal-relevant regions. Compared with existing methods on the image-goal navigation benchmark, our method brings significant performance improvement on 3 benchmark datasets (\textit{i.e.,} Gibson, MP3D, and HM3D). Especially on Gibson, we surpass the state-of-the-art success rate by 8\% with only 1/50 model size.

Network pruning and quantization are proven to be effective ways for deep model compression. To obtain a highly compact model, most methods first perform network pruning and then conduct quantization based on the pruned model. However, this strategy may ignore that the pruning and quantization would affect each other and thus performing them separately may lead to sub-optimal performance. To address this, performing pruning and quantization jointly is essential. Nevertheless, how to make a trade-off between pruning and quantization is non-trivial. Moreover, existing compression methods often rely on some pre-defined compression configurations (i.e., pruning rates or bitwidths). Some attempts have been made to search for optimal configurations, which however may take unbearable optimization cost. To address these issues, we devise a simple yet effective method named Single-path Bit Sharing (SBS) for automatic loss-aware model compression. To this end, we consider the network pruning as a special case of quantization and provide a unified view for model pruning and quantization. We then introduce a single-path model to encode all candidate compression configurations, where a high bitwidth value will be decomposed into the sum of a lowest bitwidth value and a series of re-assignment offsets. Relying on the single-path model, we introduce learnable binary gates to encode the choice of configurations and learn the binary gates and model parameters jointly. More importantly, the configuration search problem can be transformed into a subset selection problem, which helps to significantly reduce the optimization difficulty and computation cost. In this way, the compression configurations of each layer and the trade-off between pruning and quantization can be automatically determined. Extensive experiments on CIFAR-100 and ImageNet show that SBS significantly reduces computation cost while achieving promising performance. For example, our SBS compressed MobileNetV2 achieves 22.6× Bit-Operation (BOP) reduction with only 0.1% drop in the Top-1 accuracy.

Neural network quantization aims to reduce the model size, computational complexity, and memory consumption by mapping weights and activations from full-precision to low-precision. However, many existing quantization methods, either post-training with calibration or quantization-aware training with fine-tuning, require original data for better performance, which may not be available due to confidentiality or privacy constraints. This lack of data can lead to a significant decline in performance. In this paper, we propose a universal and effective method called Generative Data Free Model Quantization with Knowledge Matching for Classification(KMDFQ) that removes the dependence on data for neural network quantization. To achieve this, we propose a knowledge matching generator that produces meaningful fake data based on the latent knowledge in the pre-trained model, including classification boundary knowledge and data distribution information. Based on this generator, we propose a fake-data driven data free quantization method that uses the generated data to take advantage of the latent knowledge for quantization. Furthermore, we introduce Mean Square Error alignment during the fine-tuning of the quantized model to more strictly and directly learn knowledge, making it more suitable for data free quantization. Extensive experiments on image classification demonstrate the effectiveness of our method, achieving higher accuracy than existing data free quantization methods, particularly as the quantization bit decreases. For example, on ImageNet, the 4-bit data free quantized ResNet-18 has less than a 1.2% accuracy decline compared to quantization with real data. The source code is available athttps://github.com/ZSHsh98/KMDFQ.

Videofluoroscopic swallowing study (VFSS) visualizes the swallowing movement by using X-ray fluoroscopy, which is the most widely used method for dysphagia examination. To better facilitate swallowing assessment, the temporal parameter is one of the most important indicators. However, most information of that acquire is hand-crafted and elaborated, which is time-consuming and difficult to ensure objectivity and accuracy. In this article, we propose to formulate this task as a temporal action localization task and solve it using deep neural networks. However, the action of VFSS has the following characteristics such as small motion targets, small action amplitudes, large sample variances, short duration, and variations in duration. Furthermore, all existing methods often rely on daily behaviors, which makes locating and recognizing micro-actions more challenging. To address the above issues, we first collect and annotate the VFSS micro-action dataset, which includes 847 VFSS data from 71 subjects, due to the lack of benchmarks. We then introduce a coarse-to-fine mechanism to handle the short and repeated nature of micro-actions, which can significantly enhancing micro-action localization accuracy. Moreover, we propose a Variable-Size Window Generator method, which improves the model's characterization performance and addresses the issue of different action timings, leading to further improvements in localization accuracy. The results of our experiments demonstrate the superiority of our method, with significantly improved performance (46.10% vs. 37.70%).

We address the problem of facial motion retargeting that aims to transfer facial motion from a 2D face image to 3D characters. Existing methods often formulate this problem as a 3D face reconstruction problem, which estimates the face attributes such as face identity and expression from face images. However, due to the lack of ground-truth labels for both identity and expression, most 3D-face reconstruction-based methods fail to capture the facial identity and expression accurately. As a result, these methods may not achieve promising performance. To address this, we propose an identity-consistent constraint to learn accurate identities by encouraging consistent identity prediction across multiple frames. Based on a more accurate identity, we are able to obtain a more accurate facial expression. Moreover, we further propose an expression-exclusive constraint to improve performance by avoiding the co-occurrence of contradictory expression units (e.g., ``brow lower'' vs. ``brow raise''). Extensive experiments on facial motion retargeting and 3D face reconstruction tasks demonstrate the superiority of the proposed method over existing methods. Our code and supplementary materials are available at https://github.com/deepmo24/CPEM.

Existing Voice Cloning (VC) tasks aim to convert a para-graph text to a speech with desired voice specified by a ref-erence audio. This has significantly boosted the development of artificial speech applications. However, there also exist many scenarios that cannot be well reflected by these VC tasks, such as movie dubbing, which requires the speech to be with emotions consistent with the movie plots. To fill this gap, in this work we propose a new task named Vi-sual Voice Cloning (V2C), which seeks to convert a para-graph of text to a speech with both desired voice speci-fied by a reference audio and desired emotion specified by a reference video. To facilitate research in this field, we construct a dataset, V2C-Animation, and propose a strong baseline based on existing state-of-the-art (SoTA) VC techniques. Our dataset contains 10,217 animated movie clips covering a large variety of genres (e.g., Comedy, Fantasy) and emotions (e.g., happy, sad). We further design a set of evaluation metrics, named MCD-DTW-SL, which help eval-uate the similarity between ground-truth speeches and the synthesised ones. Extensive experimental results show that even SoTA VC methods cannot generate satisfying speeches for our V2C task. We hope the proposed new task together with the constructed dataset and evaluation metric will fa-cilitate the research in the field of voice cloning and broader vision-and-language community. Source code and dataset will be released in https://github.com/chenqi008/V2C.

Test-time adaptation provides an effective means of tackling the potential distribution shift between model training and inference, by dynamically updating the model at test time. This area has seen fast progress recently, at the effectiveness of handling test shifts. Nonetheless, prior methods still suffer two key limitations: 1) these methods rely on performing backward computation for each test sample, which takes a considerable amount of time; and 2) these methods focus on improving the performance on out-of-distribution test samples and ignore that the adaptation on test data may result in a catastrophic forgetting issue, \ie, the performance on in-distribution test samples may degrade. To address these issues, we propose an efficient anti-forgetting test-time adaptation (EATA) method. Specifically, we devise a sample-efficient entropy minimization loss to exclude uninformative samples out of backward computation, which improves the overall efficiency and meanwhile boosts the out-of-distribution accuracy. Afterward, we introduce a regularization loss to ensure that critical model weights tend to be preserved during adaptation, thereby alleviating the forgetting issue. Extensive experiments on CIFAR-10-C, ImageNet-C, and ImageNet-R verify the effectiveness and superiority of our EATA.

Getting robots to navigate to multiple objects autonomously is essential yet difficult in robot applications. One of the key challenges is how to explore environments efficiently with camera sensors only. Existing navigation methods mainly focus on fixed cameras and few attempts have been made to navigate with active cameras. As a result, the agent may take a very long time to perceive the environment due to limited camera scope. In contrast, humans typically gain a larger field of view by looking around for a better perception of the environment. How to make robots perceive the environment as efficiently as humans is a fundamental problem in robotics. In this paper, we consider navigating to multiple objects more efficiently with active cameras. Specifically, we cast moving camera to a Markov Decision Process and reformulate the active camera problem as a reinforcement learning problem. However, we have to address two new challenges: 1) how to learn a good camera policy in complex environments and 2) how to coordinate it with the navigation policy. To address these, we carefully design a reward function to encourage the agent to explore more areas by moving camera actively. Moreover, we exploit human experience to infer a rule-based camera action to guide the learning process. Last, to better coordinate two kinds of policies, the camera policy takes navigation actions into account when making camera moving decisions. Experimental results show our camera policy consistently improves the performance of multi-object navigation over four baselines on two datasets.

We address a practical yet challenging problem of training robot agents to navigate in an environment following a path described by some language instructions. The instructions often contain descriptions of objects in the environment. To achieve accurate and efficient navigation, it is critical to build a map that accurately represents both spatial location and the semantic information of the environment objects. However, enabling a robot to build a map that well represents the environment is extremely challenging as the environment often involves diverse objects with various attributes. In this paper, we propose a multi-granularity map, which contains both object fine-grained details (\eg, color, texture) and semantic classes, to represent objects more comprehensively. Moreover, we propose a weakly-supervised auxiliary task, which requires the agent to localize instruction-relevant objects on the map. Through this task, the agent not only learns to localize the instruction-relevant objects for navigation but also is encouraged to learn a better map representation that reveals object information. We then feed the learned map and instruction to a waypoint predictor to determine the next navigation goal. Experimental results show our method outperforms the state-of-the-art by 4.0% and 4.6% w.r.t. success rate both in seen and unseen environments, respectively on VLN-CE dataset. The code is available at https://github.com/PeihaoChen/WS-MGMap.

Generative adversarial networks (GANs) have shown remarkable success in generating realistic data from some predefined prior distribution (e.g., Gaussian noises). However, such prior distribution is often independent of real data and thus may lose semantic information (e.g., geometric structure or content in images) of data. In practice, the semantic information might be represented by some latent distribution learned from data. However, such latent distribution may incur difficulties in data sampling for GAN methods. In this paper, rather than sampling from the predefined prior distribution, we propose a GAN model with local coordinate coding (LCC), termed LCCGAN, to improve the performance of the image generation. First, we propose an LCC sampling method in LCCGAN to sample meaningful points from the latent manifold. With the LCC sampling method, we can explicitly exploit the local information on the latent manifold and thus produce new data with promising quality. Second, we propose an improved version, namely LCCGAN++, by introducing a higher-order term in the generator approximation. This term is able to achieve better approximation and thus further improve the performance. More critically, we derive the generalization bound for both LCCGAN and LCCGAN++ and prove that a low-dimensional input is sufficient to achieve good generalization performance. Extensive experiments on several benchmark datasets demonstrate the superiority of the proposed method over existing GAN methods.

Visual grounding (VG) aims to locate the most relevant object or region in an image, based on a natural language query. Generally, it requires the machine to first understand the query, identify the key concepts in the image, and then locate the target object by specifying its bounding box. However, in many real-world visual grounding applications, we have to face with ambiguous queries and images with complicated scene structures. Identifying the target based on highly redundant and correlated information can be very challenging, and often leading to unsatisfactory performance. To tackle this, in this paper, we exploit an attention module for each kind of information to reduce internal redundancies. We then propose an accumulated attention (A-ATT) mechanism to reason among all the attention modules jointly. In this way, the relation among different kinds of information can be explicitly captured. Moreover, to improve the performance and robustness of our VG models, we additionally introduce some noises into the training procedure to bridge the distribution gap between the human-labeled training data and the real-world poor quality data. With this "noised" training strategy, we can further learn a bounding box regressor, which can be used to refine the bounding box of the target object. We evaluate the proposed methods on four popular datasets (namely ReferCOCO, ReferCOCO+, ReferCOCOg, and GuessWhat?!). The experimental results show that our methods significantly outperform all previous works on every dataset in terms of accuracy.

Designing effective architectures is one of the key factors behind the success of deep neural networks. Existing deep architectures are either manually designed or automatically searched by some Neural Architecture Search (NAS) methods. However, even a well-designed/searched architecture may still contain many nonsignificant or redundant modules/operations (e.g., some intermediate convolution or pooling layers). Such redundancy may not only incur substantial memory consumption and computational cost but also deteriorate the performance. Thus, it is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computational cost. To this end, we have proposed a Neural Architecture Transformer (NAT) method which casts the optimization problem into a Markov Decision Process (MDP) and seeks to replace the redundant operations with more efficient operations, such as skip or null connection. Note that NAT only considers a small number of possible replacements/transitions and thus comes with a limited search space. As a result, such a small search space may hamper the performance of architecture optimization. To address this issue, we propose a Neural Architecture Transformer++ (NAT++) method which further enlarges the set of candidate transitions to improve the performance of architecture optimization. Specifically, we present a two-level transition rule to obtain valid transitions, i.e., allowing operations to have more efficient types (e.g., convolution → separable convolution) or smaller kernel sizes (e.g., 5×5 → 3×3). Note that different operations may have different valid transitions. We further propose a Binary-Masked Softmax (BMSoftmax) layer to omit the possible invalid transitions. Last, based on the MDP formulation, we apply policy gradient to learn an optimal policy, which will be used to infer the optimized architectures. Extensive experiments show that the transformed architectures significantly outperform both their original counterparts and the architectures optimized by existing methods.

We study network pruning which aims to remove redundant channels/kernels and hence speed up the inference of deep networks. Existing pruning methods either train from scratch with sparsity constraints or minimize the reconstruction error between the feature maps of the pre-trained models and the compressed ones. Both strategies suffer from some limitations: the former kind is computationally expensive and difficult to converge, while the latter kind optimizes the reconstruction error but ignores the discriminative power of channels. In this paper, we propose a simple-yet-effective method called discrimination-aware channel pruning (DCP) to choose the channels that actually contribute to the discriminative power. To this end, we first introduce additional discrimination-aware losses into the network to increase the discriminative power of the intermediate layers. Next, we select the most discriminative channels for each layer by considering the discrimination-aware loss and the reconstruction error, simultaneously. We then formulate channel pruning as a sparsity-inducing optimization problem with a convex objective and propose a greedy algorithm to solve the resultant problem. Note that a channel (3D tensor) often consists of a set of kernels (each with a 2D matrix). Besides the redundancy in channels, some kernels in a channel may also be redundant and fail to contribute to the discriminative power of the network, resulting in kernel level redundancy. To solve this issue, we propose a discrimination-aware kernel pruning (DKP) method to further compress deep networks by removing redundant kernels. To avoid manually determining the pruning rate for each layer, we propose two adaptive stopping conditions to automatically determine the number of selected channels/kernels. The proposed adaptive stopping conditions tend to yield more efficient models with better performance in practice. Extensive experiments on both image classification and face recognition demonstrate the effectiveness of our methods. For example, on ILSVRC-12, the resultant ResNet-50 model with 30 percent reduction of channels even outperforms the baseline model by 0.36 percent in terms of Top-1 accuracy. We also deploy the pruned models on a smartphone (equipped with a Qualcomm Snapdragon 845 processor). The pruned MobileNetV1 and MobileNetV2 achieve 1.93× and 1.42× inference acceleration on the mobile device, respectively, with negligible performance degradation. The source code and the pre-trained models are available at https://github.com/SCUT-AILab/DCP.

Person re-identification (Re-ID) via gait features within 3D skeleton sequences is a newly-emerging topic with several advantages. Existing solutions either rely on hand-crafted descriptors or supervised gait representation learning. This paper proposes a self-supervised gait encoding approach that can leverage unlabeled skeleton data to learn gait representations for person Re-ID. Specifically, we first create self-supervision by learning to reconstruct unlabeled skeleton sequences reversely, which involves richer high-level semantics to obtain better gait representations. Other pretext tasks are also explored to further improve self-supervised learning. Second, inspired by the fact that motion's continuity endows adjacent skeletons in one skeleton sequence and temporally consecutive skeleton sequences with higher correlations (referred as locality in 3D skeleton data), we propose a locality-aware attention mechanism and a locality-aware contrastive learning scheme, which aim to preserve locality-awareness on intra-sequence level and inter-sequence level respectively during self-supervised learning. Last, with context vectors learned by our locality-aware attention mechanism and contrastive learning scheme, a novel feature named Constrastive Attention-based Gait Encodings (CAGEs) is designed to represent gait effectively. Empirical evaluations show that our approach significantly outperforms skeleton-based counterparts by 15-40 percent Rank-1 accuracy, and it even achieves superior performance to numerous multi-modal methods with extra RGB or depth information. Our codes are available at https://github.com/Kali-Hac/Locality-Awareness-SGE.

This paper focuses on the challenging task of learning 3D object surface reconstructions from RGB images. Existing methods achieve varying degrees of success by using different surface representations. However, they all have their own drawbacks, and cannot properly reconstruct the surface shapes of complex topologies, arguably due to a lack of constraints on the topological structures in their learning frameworks. To this end, we propose to learn and use the topology-preserved, skeletal shape representation to assist the downstream task of object surface reconstruction from RGB images. Technically, we propose the novel SkeletonNet design that learns a volumetric representation of a skeleton via a bridged learning of a skeletal point set, where we use parallel decoders each responsible for the learning of points on 1D skeletal curves and 2D skeletal sheets, as well as an efficient module of globally guided subvolume synthesis for a refined, high-resolution skeletal volume; we present a differentiable Point2Voxel layer to make SkeletonNet end-to-end and trainable. With the learned skeletal volumes, we propose two models, the Skeleton-Based Graph Convolutional Neural Network (SkeGCNN) and the Skeleton-Regularized Deep Implicit Surface Network (SkeDISN), which respectively build upon and improve over the existing frameworks of explicit mesh deformation and implicit field learning for the downstream surface reconstruction task. We conduct thorough experiments that verify the efficacy of our proposed SkeletonNet. SkeGCNN and SkeDISN outperform existing methods as well, and they have their own merits when measured by different metrics. Additional results in generalized task settings further demonstrate the usefulness of our proposed methods. We have made our implementation code publicly available at https://github.com/tangjiapeng/SkeletonNet.

Temporal action localization, which requires a machine to recognize the location as well as the category of action instances in videos, has long been researched in computer vision. The main challenge of temporal action localization lies in that videos are usually long and untrimmed with diverse action contents involved. Existing state-of-the-art action localization methods divide each video into multiple action units (i.e., proposals in two-stage methods and segments in one-stage methods) and then perform action recognition/regression on each of them individually, without explicitly exploiting their relations during learning. In this paper, we claim that the relations between action units play an important role in action localization, and a more powerful action detector should not only capture the local content of each action unit but also allow a wider field of view on the context related to it. To this end, we propose a general graph convolutional module (GCM) that can be easily plugged into existing action localization methods, including two-stage and one-stage paradigms. To be specific, we first construct a graph, where each action unit is represented as a node and their relations between two action units as an edge. Here, we use two types of relations, one for capturing the temporal connections between different action units, and the other one for characterizing their semantic relationship. Particularly for the temporal connections in two-stage methods, we further explore two different kinds of edges, one connecting the overlapping action units and the other one connecting surrounding but disjointed units. Upon the graph we built, we then apply graph convolutional networks (GCNs) to model the relations among different action units, which is able to learn more informative representations to enhance action localization. Experimental results show that our GCM consistently improves the performance of existing action localization methods, including two-stage methods (e.g., CBR [15] and R-C3D [47]) and one-stage methods (e.g., D-SSAD [22]), verifying the generality and effectiveness of our GCM. Moreover, with the aid of GCM, our approach significantly outperforms the state-of-the-art on THUMOS14 (50.9 percent versus 42.8 percent). Augmentation experiments on ActivityNet also verify the efficacy of modeling the relationships between action units. The source code and the pre-trained models are available at https://github.com/Alvin-Zeng/GCM.

This paper tackles the problem of training a deep convolutional neural network of both low-bitwidth weights and activations. Optimizing a low-precision network is very challenging due to the non-differentiability of the quantizer, which may result in substantial accuracy loss. To address this, we propose three practical approaches, including (i) progressive quantization; (ii) stochastic precision; and (iii) joint knowledge distillation to improve the network training. First, for progressive quantization, we propose two schemes to progressively find good local minima. Specifically, we propose to first optimize a network with quantized weights and subsequently quantize activations. This is in contrast to the traditional methods which optimize them simultaneously. Furthermore, we propose a second progressive quantization scheme which gradually decreases the bitwidth from high-precision to low-precision during training. Second, to alleviate the excessive training burden due to the multi-round training stages, we further propose a one-stage stochastic precision strategy to randomly sample and quantize sub-networks while keeping other parts in full-precision. Finally, we adopt a novel learning scheme to jointly train a full-precision model alongside the low-precision one. By doing so, the full-precision model provides hints to guide the low-precision model training and significantly improves the performance of the low-precision network. Extensive experiments on various datasets (e.g., CIFAR-100, ImageNet) show the effectiveness of the proposed methods.

OCT fluid segmentation is a crucial task for diagnosis and therapy in ophthalmology. The current convolutional neural networks (CNNs) supervised by pixel-wise annotated masks achieve great success in OCT fluid segmentation. However, requiring pixel-wise masks from OCT images is time-consuming, expensive and expertise needed. This paper proposes an Intra- and inter-Slice Contrastive Learning Network (ISCLNet) for OCT fluid segmentation with only point supervision. Our ISCLNet learns visual representation by designing contrastive tasks that exploit the inherent similarity or dissimilarity from unlabeled OCT data. Specifically, we propose an intra-slice contrastive learning strategy to leverage the fluid-background similarity and the retinal layer-background dissimilarity. Moreover, we construct an inter-slice contrastive learning architecture to learn the similarity of adjacent OCT slices from one OCT volume. Finally, an end-to-end model combining intra- and inter-slice contrastive learning processes learns to segment fluid under the point supervision. The experimental results on two public OCT fluid segmentation datasets (i.e., AI Challenger and RETOUCH) demonstrate that the ISCLNet bridges the gap between fully-supervised and weakly-supervised OCT fluid segmentation and outperforms other well-known point-supervised segmentation methods.

Driver behavior recognition has become one of the most important tasks for intelligent vehicles. This task, however, is very challenging since the background contents in real-world driving scenarios are often very complex. More critically, the difference between driving behaviors is often very minor, making it extremely difficult to distinguish them. Existing methods often rely only on RGB frames (or skeleton data), which may fail to capture the minor differences between behaviors and appearance information of objects simultaneously and thus fail to achieve promising performance. To address the above issues, in this paper, we propose a bidirectional posture-appearance interaction network (BPAI-Net), which simultaneously considers RGB frames and skeleton (i.e., posture) data for driver behavior recognition. Specifically, we propose a posture-guided convolutional neural network (PG-CNN) and an appearance-guided graph convolutional network (AG-GCN) to extract appearance and posture features, respectively. To exploit the complementary information between appearance and posture, we use the appearance features from PG-CNN for guiding AG-GCN to exploit the contextual information (e.g., nearby objects) to enhance posture features. Then, we use the enhanced posture features from AG-GCN to help PG-CNN focus on critical local areas of video frames that are related to driver behaviors. In this sense, we are able to use the interaction between two modalities to extract more discriminative features and thus improve the recognition accuracy. Experimental results on Drive&Act dataset show that our method outperforms state-of-the-art methods by a large margin (67.83% vs. 63.64%). Furthermore, we collect a bus driver behavior recognition dataset and yield consistent performance gain against baseline methods, demonstrating the effectiveness of our method in real-world applications. The source code and trained models are available at github.com/SCUT-AILab/BPAI-Net/.

Heterogeneous Information Network (HIN) collecitve classification studies the problem of predicting labels for one type of nodes in a HIN which contains multiple types of nodes multiple types of links among them. Previous studies have revealed that exploiting relative importance of links is quite useful to improve node classification performance as connected nodes tend to have similar labels. Most existing approaches exploit the relative importance of links either by directly counting the number of connections among nodes or by learning the weight of each type of link from labeled data only. However, these approaches either neglect the importance of types of links to the class labels or may lead to overfitting problem. We propose aTensor-basedMarkov chain (T-Mark) approach, which is able to automatically and simultaneously predict the labels for unlabeled nodes and give the relative importance of types of links that actually improve the classification accuracy. Specifically, we build two tensor equations by using the HIN and features of nodes from both labeled and unlabeled data. A Markov chain-based model is proposed and it is solved by an iterative process to obtain the stationary distributions. Theoretical analyses of the existence and uniqueness of such probability distributions are given. Extensive experimental results demonstrate that T-Mark is able to achieve superior performance in the comparison and obtain reasonable relative importance of links.

Portfolio Selection is an important real-world financial task and has attracted extensive attention in artificial intelligence communities. This task, however, has two main difficulties: (i) the non-stationary price series and complex asset correlations make the learning of feature representation very hard; (ii) the practicality principle in financial markets requires controlling both transaction and risk costs. Most existing methods adopt handcraft features and/or consider no constraints for the costs, which may make them perform unsatisfactorily and fail to control both costs in practice. In this paper, we propose a cost-sensitive portfolio selection method with deep reinforcement learning. Specifically, a novel two-stream portfolio policy network is devised to extract both price series patterns and asset correlations, while a new cost-sensitive reward function is developed to maximize the accumulated return and constrain both costs via reinforcement learning. We theoretically analyze the near-optimality of the proposed reward, which shows that the growth rate of the policy regarding this reward function can approach the theoretical optimum. We also empirically evaluate the proposed method on real-world datasets. Promising results demonstrate the effectiveness and superiority of the proposed method in terms of profitability, cost-sensitivity and representation abilities.

View synthesis aims to produce unseen views from a set of views captured by two or more cameras at different positions. This task is non-trivial since it is hard to conduct pixel-level matching among different views. To address this issue, most existing methods seek to exploit the geometric information to match pixels. However, when the distinct cameras have a large baseline (i. e., far away from each other), severe geometry distortion issues would occur and the geometric information may fail to provide useful guidance, resulting in very blurry synthesized images. To address the above issues, in this paper, we propose a novel deep generative model, called Self-Consistent Generative Network (SCGN), which synthesizes novel views from the given input views without explicitly exploiting the geometric information. The proposed SCGN model consists of two main components, i. e., a View Synthesis Network (VSN) and a View Decomposition Network (VDN), both employing an Encoder-Decoder structure. Here, the VDN seeks to reconstruct input views from the synthesized novel view to preserve the consistency of view synthesis. Thanks to VDN, SCGN is able to synthesize novel views without using any geometric rectification before encoding, making it easier for both training and applications. Finally, adversarial loss is introduced to improve the photo-realism of novel views. Both qualitative and quantitative comparisons against several state-of-the-art methods on two benchmark tasks demonstrated the superiority of our approach.

Computation partitioning is an important technique to improve the application performance by selectively offloading some computations from the mobile devices to the nearby edge cloud. In a dynamic environment in which the network bandwidth to the edge cloud may change frequently, the partitioning of the computation needs to be updated accordingly. The frequent updating of partitioning leads to high state migration cost between the mobile side and edge cloud. However, existing works don’t take the state migration overhead into consideration. Consequently, the partitioning decisions may cause significant network congestion and increase overall completion time tremendously. In this article, with considering the state migration overhead, we propose a set of novel algorithms to update the partitioning based on the changing network bandwidth. To the best of our knowledge, this is the first work on computation partitioning for stateful data stream applications in dynamic environments. The algorithms aim to alleviate the network congestion and minimize the make-span through selectively migrating state in dynamic edge cloud environments. Extensive simulations show our solution not only could selectively migrate state but also outperforms other classical benchmark algorithms in terms of make-span. The proposed model and algorithms will enrich the scheduling theory forstatefultasks, which has not been explored before.

We study unsupervised video representation learning that seeks to learn both motion and appearance features from unlabeled video only, which can be reused for downstream tasks such as action recognition. This task, however, is extremely challenging due to 1) the highly complex spatial-temporal information in videos and 2) the lack of labeled data for training. Unlike representation learning for static images, it is difficult to construct a suitable self-supervised task to effectively model both motion and appearance features. More recently, several attempts have been made to learn video representation through video playback speed prediction. However, it is non-trivial to obtain precise speed labels for the videos. More critically, the learned models may tend to focus on motion patterns and thus may not learn appearance features well. In this paper, we observe that the relative playback speed is more consistent with motion patterns and thus provides more effective and stable supervision for representation learning. Therefore, we propose a new way to perceive the playback speed and exploit the relative speed between two video clips as labels. In this way, we are able to effectively perceive speed and learn better motion features. Moreover, to ensure the learning of appearance features, we further propose an appearance-focused task, where we enforce the model to perceive the appearance difference between two video clips. We show that jointly optimizing the two tasks consistently improves the performance on two downstream tasks (namely, action recognition and video retrieval) w.r.t the increasing pre-training epochs. Remarkably, for action recognition on the UCF101 dataset, we achieve 93.7% accuracy without the use of labeled data for pre-training, which outperforms the ImageNet supervised pre-trained model. Our code, pre-trained models, and supplementary materials can be found at https://github.com/PeihaoChen/RSPNet.

Reading and writing research papers is one of the most privileged abilities that a qualified researcher should master. However, it is difficult for new researchers (e.g., students) to fully grasp this ability. It would be fascinating if we could train an intelligent agent to help people read and summarize papers, and perhaps even discover and exploit the potential knowledge clues to write novel papers. Although there have been existing works focusing on summarizing (i.e., reading) the knowledge in a given text or generating (i.e., writing) a text based on the given knowledge, the ability of simultaneously reading and writing is still under development. Typically, this requires an agent to fully understand the knowledge from the given text materials and generate correct and fluent novel paragraphs, which is very challenging in practice. In this paper, we propose a Deep ReAder-Writer (DRAW) network, which consists of a Reader that can extract knowledge graphs (KGs) from input paragraphs and discover potential knowledge, a graph-to-text Writer that generates a novel paragraph, and a Reviewer that reviews the generated paragraph from three different aspects. Extensive experiments show that our DRAW network outperforms considered baselines and several state-of-the-art methods on AGENDA and M-AGENDA datasets. Our code and supplementary are released at https://github.com/menggehe/DRAW.

One of the key steps in Neural Architecture Search (NAS) is to estimate the performance of candidate architectures. Existing methods either directly use the validation performance or learn a predictor to estimate the performance. However, these methods can be either computationally expensive or very inaccurate, which may severely affect the search efficiency and performance. Moreover, as it is very difficult to annotate architectures with accurate performance on specific tasks, learning a promising performance predictor is often non-trivial due to the lack of labeled data. In this paper, we argue that it may not be necessary to estimate the absolute performance for NAS. On the contrary, we may need only to understand whether an architecture is better than a baseline one. However, how to exploit this comparison information as the reward and how to well use the limited labeled data remains two great challenges. In this paper, we propose a novel Contrastive Neural Architecture Search (CTNAS) method which performs architecture search by taking the comparison results between architectures as the reward. Specifically, we design and learn a Neural Architecture Comparator (NAC) to compute the probability of candidate architectures being better than a baseline one. Moreover, we present a baseline updating scheme to improve the baseline iteratively in a curriculum learning manner. More critically, we theoretically show that learning NAC is equivalent to optimizing the ranking over architectures. Extensive experiments in three search spaces demonstrate the superiority of our CTNAS over existing methods.

Network quantization allows inference to be conducted using low-precision arithmetic for improved inference efficiency of deep neural networks on edge devices. However, designing aggressively low-bit (e.g., 2-bit) quantization schemes on complex tasks, such as object detection, still remains challenging in terms of severe performance degradation and unverifiable efficiency on common hardware. In this paper, we propose an Accurate Quantized object Detection solution, termed AQD, to fully get rid of floating-point computation. To this end, we target using fixed-point operations in all kinds of layers, including the convolutional layers, normalization layers, and skip connections, allowing the inference to be executed using integer-only arithmetic. To demonstrate the improved latency-vs-accuracy trade-off, we apply the proposed methods on RetinaNet and FCOS. In particular, experimental results on MS-COCO dataset show that our AQD achieves comparable or even better performance compared with the full-precision counterpart under extremely low-bit schemes, which is of great practical value. Source code and models are available at: https://github.com/aim-uofa/model-quantization

Text-based image captioning (TextCap) which aims to read and reason images with texts is crucial for a machine to understand a detailed and complex scene environment, considering that texts are omnipresent in daily life. This task, however, is very challenging because an image often contains complex texts and visual information that is hard to be described comprehensively. Existing methods attempt to extend the traditional image captioning methods to solve this task, which focus on describing the overall scene of images by one global caption. This is infeasible because the complex text and visual information cannot be described well within one caption. To resolve this difficulty, we seek to generate multiple captions that accurately describe different parts of an image in detail. To achieve this purpose, there are three key challenges: 1) it is hard to decide which parts of the texts of images to copy or paraphrase; 2) it is non-trivial to capture the complex relationship between diverse texts in an image; 3) how to generate multiple captions with diverse content is still an open problem. To conquer these, we propose a novel Anchor-Captioner method. Specifically, we first find the important tokens which are supposed to be paid more attention to and consider them as anchors. Then, for each chosen anchor, we group its rel-evant texts to construct the corresponding anchor-centred graph (ACG). Last, based on different ACGs, we conduct the multi-view caption generation to improve the content diversity of generated captions. Experimental results show that our method not only achieves SOTA performance but also generates diverse captions to describe images.

We study self-supervised video representation learning, which is a challenging task due to 1) lack of labels for explicit supervision; 2) unstructured and noisy visual information. Existing methods mainly use contrastive loss with video clips as the instances and learn visual representation by discriminating instances from each other, but they need a careful treatment of negative pairs by either relying on large batch sizes, memory banks, extra modalities or customized mining strategies, which inevitably includes noisy data. In this paper, we observe that the consistency between positive samples is the key to learn robust video representation. Specifically, we propose two tasks to learn appearance and speed consistency, respectively. The appearance consistency task aims to maximize the similarity between two clips of the same video with different playback speeds. The speed consistency task aims to maximize the similarity between two clips with the same playback speed but different appearance information. We show that optimizing the two tasks jointly consistently improves the performance on downstream tasks, e.g., action recognition and video retrieval. Remarkably, for action recognition on the UCF-101 dataset, we achieve 90.8% accuracy without using any extra modalities or negative pairs for unsupervised pretraining, which outperforms the ImageNet supervised pretrained model. Codes and models will be available.

Instance segmentation in 3D scenes is fundamental in many applications of scene understanding. It is yet challenging due to the compound factors of data irregularity and uncertainty in the numbers of instances. State-of-the-art methods largely rely on a general pipeline that first learns point-wise features discriminative at semantic and instance levels, followed by a separate step of point grouping for proposing object instances. While promising, they have the shortcomings that (1) the second step is not supervised by the main objective of instance segmentation, and (2) their point-wise feature learning and grouping are less effective to deal with data irregularities, possibly resulting in fragmented segmentations. To address these issues, we propose in this work an end-to-end solution of Semantic Superpoint Tree Network (SSTNet) for proposing object instances from scene points. Key in SST-Net is an intermediate, semantic superpoint tree (SST), which is constructed based on the learned semantic features of superpoints, and which will be traversed and split at intermediate tree nodes for proposals of object instances. We also design in SSTNet a refinement module, termed CliqueNet, to prune superpoints that may be wrongly grouped into instance proposals. Experiments on the benchmarks of ScanNet and S3DIS show the efficacy of our proposed method. At the time of submission, SSTNet ranks top on the ScanNet (V2) leaderboard, with 2% higher of mAP than the second best method. The source code in PyTorch is available at https://github.com/Gorilla-Lab-SCUT/SSTNet.

3D LiDAR (light detection and ranging) semantic segmentation is important in scene understanding for many applications, such as auto-driving and robotics. For example, for autonomous cars equipped with RGB cameras and LiDAR, it is crucial to fuse complementary information from different sensors for robust and accurate segmentation. Existing fusion-based methods, however, may not achieve promising performance due to the vast difference between the two modalities. In this work, we investigate a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds. To this end, we first project point clouds to the camera coordinates to provide spatio-depth information for RGB images. Then, we propose a two-stream network to extract features from the two modalities, separately, and fuse the features by effective residual-based fusion modules. Moreover, we propose additional perception-aware losses to measure the perceptual difference between the two modalities. Extensive experiments on two benchmark data sets show the superiority of our method. For example, on nuScenes, our PMF outperforms the state-of-the-art method by 0.8% in mIoU.

In real-world applications, data often come in a growing manner, where the data volume and the number of classes may increase dynamically. This will bring a critical challenge for learning: given the increasing data volume or the number of classes, one has to instantaneously adjust the neural model capacity to obtain promising performance. Existing methods either ignore the growing nature of data or seek to independently search an optimal architecture for a given dataset, and thus are incapable of promptly adjusting the architectures for the changed data. To address this, we present a neural architecture adaptation method, namely Adaptation eXpert (AdaXpert), to efficiently adjust previous architectures on the growing data. Specifically, we introduce an architecture adjuster to generate a suitable architecture for each data snapshot, based on the previous architecture and the different extent between current and previous data distributions. Furthermore, we propose an adaptation condition to determine the necessity of adjustment, thereby avoiding unnecessary and time-consuming adjustments. Extensive experiments on two growth scenarios (increasing data volume and number of classes) demonstrate the effectiveness of the proposed method.

We study a practical domain adaptation task, called source-free unsupervised domain adaptation (UDA) problem, in which we cannot access source domain data due to data privacy issues but only a pre-trained source model and unlabeled target data are available. This task, however, is very difficult due to one key challenge: the lack of source data and target domain labels makes model adaptation very challenging. To address this, we propose to mine the hidden knowledge in the source model and exploit it to generate source avatar prototypes (i.e. representative features for each source class) as well as target pseudo labels for domain alignment. To this end, we propose a Contrastive Prototype Generation and Adaptation (CPGA) method. Specifically, CPGA consists of two stages: (1) prototype generation: by exploring the classification boundary information of the source model, we train a prototype generator to generate avatar prototypes via contrastive learning. (2) prototype adaptation: based on the generated source prototypes and target pseudo labels, we develop a new robust contrastive prototype adaptation strategy to align each pseudo-labeled target data to the corresponding source prototypes. Extensive experiments on three UDA benchmark datasets demonstrate the effectiveness and superiority of the proposed method.

Mathematical expression recognition (MER) aims to convert an image of mathematical expressions into a Latex sequence. In practice, the task of MER is challenging, since 1) the images of mathematical expressions often contain complex structure relationships, e.g., fractions, matrixes, and subscripts; 2) the generated Latex sequences can be very complex and they have to satisfy strict syntax rules. Existing methods, however, often ignore the complex dependence among image regions, resulting in poor feature representation. In addition, they may fail to capture the rigorous relations among different formula symbols as they consider MER as a common language generation task. To address these issues, we propose a Structure-Aware Sequence-Level (SASL) model for MER. First, to better represent and recognize the visual content of formula images, we propose a structure-aware module to capture the relationship among different symbols. Meanwhile, the sequence-level modeling helps the model to concentrate on the generation of entire sequences. To make the problem feasible, we cast the generation problem into a Markov decision process (MDP) and seek to learn a Latex sequence generating policy. Based on MDP, we learn SASL by maximizing the matching score of each image-sequence pair to obtain the generation policy. Extensive experiments on the IM2LATEX-100K dataset verify the effectiveness and superiority of the proposed method.

Despite recent significant progress on generative models, context-rich text-to-image synthesis depicting multiple complex objects is still non-trivial. The main challenges lie in the ambiguous semantic of a complex description and the intricate scene of an image with various objects, different positional relationship and diverse appearances. To address these challenges, we propose R-GAN, which can generate reasonable images according to the given text in a human-like way. Specifically, just like humans will first find and settle the essential elements to create a simple sketch, we first capture a monolithic-structural text representation by building a scene graph to find the essential semantic elements. Then, based on this representation, we design a bounding box generator to estimate the layout with position and size of target objects, and a following shape generator, which draws a fine-detailed shape for each object. Different from previous work only generating coarse shapes blindly, we introduce a coarse-to-fine shape generator based on a shape knowledge base. At last, to finish the final image synthesis, we propose a multi-modal geometry-aware spatially-adaptive generator conditioned on the monolithic-structural text representation and the geometry-aware map of the shapes. Extensive experiments on the real-world dataset MSCOCO show the superiority of our method in terms of both quantitative and qualitative metrics.

Visual question answering (VQA) is designed to examine the visual-textual reasoning ability of an intelligent agent. However, recent observations show that many VQA models may only capture the biases between questions and answers in a dataset rather than showing real reasoning abilities. For example, given a question, some VQA models tend to output the answer that occurs frequently in the dataset and ignore the images. To reduce this tendency, existing methods focus on weakening the language bias. Meanwhile, only a few works also consider vision bias implicitly. However, these methods introduce additional annotations or show unsatisfactory performance. Moreover, not all biases are harmful to the models. Some “biases” learnt from datasets represent natural rules of the world and can help limit the range of answers. Thus, how to filter and remove the true negative biases in language and vision modalities remain a major challenge. In this paper, we propose a method named D-VQA to alleviate the above challenges from the feature and sample perspectives. Specifically, from the feature perspective, we build a question-to-answer and vision-to-answer branch to capture the language and vision biases, respectively. Next, we apply two unimodal bias detection modules to explicitly recognise and remove the negative biases. From the sample perspective, we construct two types of negative samples to assist the training of the models, without introducing additional annotations. Extensive experiments on the VQA-CP v2 and VQA v2 datasets demonstrate the effectiveness of our D-VQA method.

High-resolution representations are essential for position-sensitive vision problems, such as human pose estimation, semantic segmentation, and object detection. Existing state-of-the-art frameworks first encode the input image as a low-resolution representation through a subnetwork that is formed by connecting high-to-low resolution convolutions in series (e.g., ResNet, VGGNet), and then recover the high-resolution representation from the encoded low-resolution representation. Instead, our proposed network, named as High-Resolution Network (HRNet), maintains high-resolution representations through the whole process. There are two key characteristics: (i) Connect the high-to-low resolution convolution streams in parallel and (ii) repeatedly exchange the information across resolutions. The benefit is that the resulting representation is semantically richer and spatially more precise. We show the superiority of the proposed HRNet in a wide range of applications, including human pose estimation, semantic segmentation, and object detection, suggesting that the HRNet is a stronger backbone for computer vision problems. All the codes are available at https://github.com/HRNet.

Deep neural networks have achieved remarkable success for video-based action recognition. However, most of existing approaches cannot be deployed in practice due to the high computational cost. To address this challenge, we propose a new real-time convolutional architecture, called Temporal Convolutional 3D Network (T-C3D), for action representation. T-C3D learns video action representations in a hierarchical multi-granularity manner while obtaining a high process speed. Specifically, we propose a residual 3D Convolutional Neural Network (CNN) to capture complementary information on the appearance of a single frame and the motion between consecutive frames. Based on this CNN, we develop a new temporal encoding method to explore the temporal dynamics of the whole video. Furthermore, we integrate deep compression techniques with T-C3D to further accelerate the deployment of models via reducing the size of the model. By these means, heavy calculations can be avoided when doing the inference, which enables the method to deal with videos beyond real-time speed while keeping promising performance. We validate our approach by studying its action representation performance on four benchmarks over three different tasks. Our method achieves clear improvements on UCF101 action recognition benchmark against the state-of-the-art real-time methods by 5.4% in terms of accuracy and 2 times faster in terms of inference speed with a less than 5MB storage model. The source code and the pre-trained models are publicly available at https://github.com/tc3d.

Sparse principal component analysis (SPCA) produces principal components with sparse loadings, which is very important for handling data with many irrelevant features and also critical to interpret the results. To deal with orthogonal constraints, most previous approaches address SPCA with several components using techniques such as deflation technique and convex relaxations. However, the deflation technique usually suffers from suboptimal solutions due to poor approximations. On the other hand, the convex relaxations are often computationally expensive. To address the above issues, in this paper, we propose to address SPCA over the Stiefel manifold directly, and develop a stabilized Alternating Direction Method of Multipliers (SADMM) to handle the nonconvex orthogonal constraints. Compared to traditional ADMM, the proposed SADMM method converges well with a wide range of parameters and obtains a better solution. We also theoretically study the convergence property of the proposed SADMM method. Furthermore, most existing methods ignore an inherent drawback of SPCA - the importance of different components is not considered when doing feature selection, which often makes the selected features nonoptimal. To address this, we further propose a two-stage method which considers the importance of different components to select the most important features. Empirical studies on both synthetic and real-world datasets show that the proposed algorithms achieve better performance compared to existing state-of-the-art methods.

Online Active Learning (OAL) aims to manage unlabeled datastream by selectively querying the label of data. OAL is applicable to many real-world problems, such as anomaly detection in health-care and finance. In these problems, there are two key challenges: the query budget is often limited; the ratio between classes is highly imbalanced. In practice, it is quite difficult to handle imbalanced unlabeled datastream when only a limited budget of labels can be queried for training. To solve this, previous OAL studies adopt either asymmetric losses or queries (an isolated asymmetric strategy) to tackle the imbalance, and use first-order methods to optimize the cost-sensitive measure. However, the isolated strategy limits their performance in class imbalance, while first-order methods restrict their optimization performance. In this article, we propose a novel Online Adaptive Asymmetric Active learning algorithm, based on a new asymmetric strategy (merging both asymmetric losses and queries strategies), and second-order optimization. We theoretically analyze its mistake bound and cost-sensitive metric bounds. Moreover, to better balance performance and efficiency, we enhance our algorithm via a sketching technique, which significantly accelerates the computational speed with quite slight performance degradation. Promising results demonstrate the effectiveness and efficiency of the proposed methods.

We addressed the challenging task of video question answering, which requires machines to answer questions about videos in a natural language form. Previous state-of-the-art methods attempt to apply spatio-temporal attention mechanism on video frame features without explicitly modeling the location and relations among object interaction occurred in videos. However, the relations between object interaction and their location information are very critical for both action recognition and question reasoning. In this work, we propose to represent the contents in the video as a location-aware graph by incorporating the location information of an object into the graph construction. Here, each node is associated with an object represented by its appearance and location features. Based on the constructed graph, we propose to use graph convolution to infer both the category and temporal locations of an action. As the graph is built on objects, our method is able to focus on the foreground action contents for better video question answering. Lastly, we leverage an attention mechanism to combine the output of graph convolution and encoded question features for final answer reasoning. Extensive experiments demonstrate the effectiveness of the proposed methods. Specifically, our method significantly outperforms state-of-the-art methods on TGIF-QA, Youtube2Text-QA and MSVD-QA datasets.

Low bit-width model quantization is highly desirable when deploying a deep neural network on mobile and edge devices. Quantization is an effective way to reduce the model size with low bit-width weight representation. However, the unacceptable accuracy drop hinders the development of this approach. One possible reason for this is that the weights in quantization intervals are directly assigned to the center. At the same time, some quantization applications are limited by the various of different network models. Accordingly, in this paper, we propose Multiple Phase Adaptations (MPA), a framework designed to address these two problems. Firstly, weights in the target interval are assigned to center by gradually spreading the quantization range. During the MPA process, the accuracy drop can be compensated for the unquantized parts. Moreover, as MPA does not introduce hyperparameters that depend on different models or bit-width, the framework can be conveniently applied to various models. Extensive experiments demonstrate that MPA achieves higher accuracy than most existing methods on classification tasks for AlexNet, VGG-16 and ResNet.

Home design is a complex task that normally requires architects to finish with their professional skills and tools. It will be fascinating that if one can produce a house plan intuitively without knowing much knowledge about home design and experience of using complex designing tools, for example, via natural language. In this paper, we formulate it as a language conditioned visual content generation problem that is further divided into a floor plan generation and an interior texture (such as floor and wall) synthesis task. The only control signal of the generation process is the linguistic expression given by users that describe the house details. To this end, we propose a House Plan Generative Model (HPGM) that first translates the language input to a structural graph representation and then predicts the layout of rooms with a Graph Conditioned Layout Prediction Network (GC-LPN) and generates the interior texture with a Language Conditioned Texture GAN (LCT-GAN). With some post-processing, the final product of this task is a 3D house model. To train and evaluate our model, we build the first Text-to-3D House Model dataset, which will be released at: https:// hidden-link-for-submission.

Deep neural networks have exhibited promising performance in image super-resolution (SR) by learning a nonlinear mapping function from low-resolution (LR) images to high-resolution (HR) images. However, there are two underlying limitations to existing SR methods. First, learning the mapping function from LR to HR images is typically an ill-posed problem, because there exist infinite HR images that can be downsampled to the same LR image. As a result, the space of the possible functions can be extremely large, which makes it hard to find a good solution. Second, the paired LR-HR data may be unavailable in real-world applications and the underlying degradation method is often unknown. For such a more general case, existing SR models often incur the adaptation problem and yield poor performance. To address the above issues, we propose a dual regression scheme by introducing an additional constraint on LR data to reduce the space of the possible functions. Specifically, besides the mapping from LR to HR images, we learn an additional dual regression mapping estimates the down-sampling kernel and reconstruct LR images, which forms a closed-loop to provide additional supervision. More critically, since the dual regression process does not depend on HR images, we can directly learn from LR images. In this sense, we can easily adapt SR models to real-world data, e.g., raw video frames from YouTube. Extensive experiments with paired training data and unpaired real-world data demonstrate our superiority over existing methods.

We address the problem of video grounding from natural language queries. The key challenge in this task is that one training video might only contain a few annotated starting/ending frames that can be used as positive examples for model training. Most conventional approaches directly train a binary classifier using such imbalance data, thus achieving inferior results. The key idea of this paper is to use the distances between the frame within the ground truth and the starting (ending) frame as dense supervisions to improve the video grounding accuracy. Specifically, we design a novel dense regression network (DRN) to regress the distances from each frame to the starting (ending) frame of the video segment described by the query. We also propose a simple but effective IoU regression head module to explicitly consider the localization quality of the grounding results (i.e., the IoU between the predicted location and the ground truth). Experimental results show that our approach significantly outperforms state-of-the-arts on three datasets (i.e., Charades-STA, ActivityNet-Captions, and TACoS).

In this paper, we seek to tackle a challenge in training low-precision networks: the notorious difficulty in propagating gradient through a low-precision network due to the non-differentiable quantization function. We propose a solution by training the low-precision network with a full-precision auxiliary module. Specifically, during training, we construct a mix-precision network by augmenting the original low-precision network with the full precision auxiliary module. Then the augmented mix-precision network and the low-precision network are jointly optimized. This strategy creates additional full-precision routes to update the parameters of the low-precision model, thus making the gradient back-propagates more easily. At the inference time, we discard the auxiliary module without introducing any computational complexity to the low-precision network. We evaluate the proposed method on image classification and object detection over various quantization approaches and show consistent performance increase. In particular, we achieve near lossless performance to the full-precision model by using a 4-bit detector, which is of great practical value.

Neural architecture search (NAS) has become an important approach to automatically find effective architectures. To cover all possible good architectures, we need to search in an extremely large search space with billions of candidate architectures. More critically, given a large search space, we may face a very challenging issue of space explosion. However, due to the limitation of computational resources, we can only sample a very small proportion of the architectures, which provides insufficient information for the training. As a result, existing methods may often produce sub-optimal architectures. To alleviate this issue, we propose a curriculum search method that starts from a small search space and gradually incorporates the learned knowledge to guide the search in a large space. With the proposed search strategy, our Curriculum Neural Architecture Search (CNAS) method significantly improves the search efficiency and finds better architectures than existing NAS methods. Extensive experiments on CIFAR-10 and ImageNet demonstrate the effectiveness of the proposed method.

Gait-based person re-identification (Re-ID) is valuable for safety-critical applications, and using only 3D skeleton data to extract discriminative gait features for person Re-ID is an emerging open topic. Existing methods either adopt hand-crafted features or learn gait features by traditional supervised learning paradigms. Unlike previous methods, we for the first time propose a generic gait encoding approach that can utilize unlabeled skeleton data to learn gait representations in a self-supervised manner. Specifically, we first propose to introduce self-supervision by learning to reconstruct input skeleton sequences in reverse order, which facilitates learning richer high-level semantics and better gait representations. Second, inspired by the fact that motion's continuity endows temporally adjacent skeletons with higher correlations (“locality”), we propose a locality-aware attention mechanism that encourages learning larger attention weights for temporally adjacent skeletons when reconstructing current skeleton, so as to learn locality when encoding gait. Finally, we propose Attention-based Gait Encodings (AGEs), which are built using context vectors learned by locality-aware attention, as final gait representations. AGEs are directly utilized to realize effective person Re-ID. Our approach typically improves existing skeleton-based methods by 10-20% Rank-1 accuracy, and it achieves comparable or even superior performance to multi-modal methods with extra RGB or depth information.

Portfolio selection is an important yet challenging task in AI for FinTech. One of the key issues is how to represent the non-stationary price series of assets in a portfolio, which is important for portfolio decisions. The existing methods, however, fall short of capturing: 1) the complicated sequential patterns for asset price series and 2) the price correlations among multiple assets. In this paper, under a deep reinforcement learning paradigm for portfolio selection, we propose a novel Relation-aware Transformer (RAT) to handle these aspects. Specifically, being equipped with our newly developed attention modules, RAT is structurally innovated to capture both sequential patterns and asset correlations for portfolio selection. Based on the extracted sequential features, RAT is able to make profitable portfolio decisions regarding each asset via a newly devised leverage operation. Extensive experiments on real-world crypto-currency and stock datasets verify the state-of-the-art performance of RAT.

Semantic segmentation requires a large amount of densely annotated data for training and may generalize poorly to novel categories. In real-world applications, we have an urgent need for few-shot semantic segmentation which aims to empower a model to handle unseen object categories with limited data. This task is non-trivial due to several challenges. First, it is difficult to extract the class-relevant information to handle the novel class as only a few samples are available. Second, since the image content can be very complex, the novel class information may be suppressed by the base categories due to limited data. Third, one may easily learn promising base classifiers based on a large amount of training data, but it is non-trivial to exploit the knowledge to train the novel classifiers. More critically, once a novel classifier is built, the output probability space will change. How to maintain the base classifiers and dynamically include the novel classifiers remains an open question. To address the above issues, we propose a Dynamic Extension Network (DENet) in which we dynamically construct and maintain a classifier for the novel class by leveraging the knowledge from the base classes and the information from novel data. More importantly, to overcome the information suppression issue, we design a Guided Attention Module (GAM), which can be plugged into any framework to help learn class-relevant features. Last, rather than directly train the model with limited data, we propose a dynamic extension training algorithm to predict the weights of novel classifiers, which is able to exploit the knowledge of base classifiers by dynamically extending classes during training. The extensive experiments show that our proposed method achieves state-of-the-art performance on the PASCAL-5i and COCO-20i datasets. The source code is available at https://github.com/lizhaoliu-Lec/DENet.

We study the problem of text-based visual question answering (T-VQA) in this paper. Unlike general visual question answering (VQA) which only builds connections between questions and visual contents, T-VQA requires reading and reasoning over both texts and visual concepts that appear in images. Challenges in T-VQA mainly lie in three aspects: 1) It is difficult to understand the complex logic in questions and extract specific useful information from rich image contents to answer them; 2) The text-related questions are also related to visual concepts, but it is difficult to capture cross-modal relationships between the texts and the visual concepts; 3) If the OCR (optical character recognition) system fails to detect the target text, the training will be very difficult. To address these issues, we propose a novel Cascade Reasoning Network (CRN) that consists of a progressive attention module (PAM) and a multimodal reasoning graph (MRG) module. Specifically, the PAM regards the multimodal information fusion operation as a stepwise encoding process and uses the previous attention results to guide the next fusion process. The MRG aims to explicitly model the connections and interactions between texts and visual concepts. To alleviate the dependence on the OCR system, we introduce an auxiliary task to train the model with accurate supervision signals, thereby enhancing the reasoning ability of the model in question answering. Extensive experiments on three popular T-VQA datasets demonstrate the effectiveness of our method compared with SOTA methods. The source code is available at https://github.com/guanghuixu/CRN_tvqa.

We address the challenging task of event localization, which requires the machine to localize an event and recognize its category in unconstrained videos. Most existing methods leverage only the visual information of a video while neglecting its audio information, which, however, can be very helpful and important for event localization. For example, humans often recognize an event by reasoning with the visual and audio content simultaneously. Moreover, the audio information can guide the model to pay more attention on the informative regions of visual scenes, which can help to reduce the interference brought by the background. Motivated by these, in this paper, we propose a relation-aware network to leverage both audio and visual information for accurate event localization. Specifically, to reduce the interference brought by the background, we propose an audio-guided spatial-channel attention module to guide the model to focus on event-relevant visual regions. Besides, we propose to build connections between visual and audio modalities with a relation-aware module. In particular, we learn the representations of video and/or audio segments by aggregating information from the other modality according to the cross-modal relations. Last, relying on the relation-aware representations, we conduct event localization by predicting the event relevant score and classification score. Extensive experimental results demonstrate that our method significantly outperforms the state-of-the-arts in both supervised and weakly-supervised AVE settings.

Generating videos given a text description (such as a script) is non-trivial due to the intrinsic complexity of image frames and the structure of videos. Although Generative Adversarial Networks (GANs) have been successfully applied to generate images conditioned on a natural language description, it is still very challenging to generate realistic videos in which the frames are required to follow both spatial and temporal coherence. In this paper, we propose a novel Bottom-up GAN (BoGAN) method for generating videos given a text description. To ensure the coherence of the generated frames and also make the whole video match the language descriptions semantically, we design a bottom-up optimisation mechanism to train BoGAN. Specifically, we devise a region-level loss via attention mechanism to preserve the local semantic alignment and draw details in different sub-regions of video conditioned on words which are most relevant to them. Moreover, to guarantee the matching between text and frame, we introduce a frame-level discriminator, which can also maintain the fidelity of each frame and the coherence across frames. Last, to ensure the global semantic alignment between whole video and given text, we apply a video-level discriminator. We evaluate the effectiveness of the proposed BoGAN on two synthetic datasets (i.e., SBMG and TBMG) and two real-world datasets (i.e., MSVD and KTH).

We focus on the task of generating sound from natural videos, and the sound should be both temporally and content-wise aligned with visual signals. This task is extremely challenging because some sounds generated outside a camera can not be inferred from video content. The model may be forced to learn an incorrect mapping between visual content and these irrelevant sounds. To address this challenge, we propose a framework named REGNET. In this framework, we first extract appearance and motion features from video frames to better distinguish the object that emits sound from complex background information. We then introduce an innovative audio forwarding regularizer that directly considers the real sound as input and outputs bottlenecked sound features. Using both visual and bottlenecked sound features for sound prediction during training provides stronger supervision for the sound prediction. The audio forwarding regularizer can control the irrelevant sound component and thus prevent the model from learning an incorrect mapping between video frames and sound emitted by the object that is out of the screen. During testing, the audio forwarding regularizer is removed to ensure that REGNET can produce purely aligned sound only from visual features. Extensive evaluations based on Amazon Mechanical Turk demonstrate that our method significantly improves both temporal and contentwise alignment. Remarkably, our generated sound can fool the human with a 68.12% success rate. Code and pre-trained models are publicly available at https://github.com/PeihaoChen/regnet.

Deep learning based medical image diagnosis has shown great potential in clinical medicine. However, it often suffers two major difficulties in real-world applications: 1) only limited labels are available for model training, due to expensive annotation costs over medical images; 2) labeled images may contain considerable label noise (e.g.,mislabeling labels) due to diagnostic difficulties of diseases. To address these, we seek to exploit rich labeled data from relevant domains to help the learning in the target task via Unsupervised Domain Adaptation (UDA). Unlike most UDA methods that rely on clean labeled data or assume samples are equally transferable, we innovatively propose a Collaborative Unsupervised Domain Adaptation algorithm, which conducts transferability-aware adaptation and conquers label noise in a collaborative way. We theoretically analyze the generalization performance of the proposed method, and also empirically evaluate it on both medical and general images. Promising experimental results demonstrate the superiority and generalization of the proposed method.

Spectral analysis is critical in social network analysis. As a vital step of the spectral analysis, the graph construction in many existing works utilizes content data only. Unfortunately, the content data often consists of noisy, sparse, and redundant features, which makes the resulting graph unstable and unreliable. In practice, besides the content data, social network data also contain link information, which provides additional information for graph construction. Some of previous works utilize the link data. However, the link data is often incomplete, which makes the resulting graph incomplete. To address these issues, we propose a novel Distilled Graph Clustering (DGC) method. It pursuits adistilled graphbased on both the content data and the link data. The proposed algorithm alternates between two steps: in the feature selection step, it finds the most representative feature subset w.r.t. an intermediate graph initialized with link data; in graph distillation step, the proposed method updates and refines the graph based on only the selected features. The final resulting graph, which is referred to as the distilled graph, is then utilized for spectral clustering on the large-scale social network data. Extensive experiments demonstrate the superiority of the proposed method.

Matrix factorization has been widely applied to various applications. With the fast development of storage and internet technologies, we have been witnessing a rapid increase of data. In this paper, we propose new algorithms for matrix factorization with the emphasis on efficiency. In addition, most existing methods of matrix factorization only consider a general smooth least square loss. Differently, many real-world applications have distinctive characteristics. As a result, different losses should be used accordingly. Therefore, it is beneficial to design new matrix factorization algorithms that are able to deal with both smooth and non-smooth losses. To this end, one needs to analyze the characteristics of target data and use the most appropriate loss based on the analysis. We particularly study two representative cases of low-rank matrix recovery, i.e., collaborative filtering for recommendation and high dynamic range imaging. To solve these two problems, we respectively propose a stage-wise matrix factorization algorithm by exploiting manifold optimization techniques. From our theoretical analysis, they are both are provably guaranteed to converge to a stationary point. Extensive experiments on recommender systems and high dynamic range imaging demonstrate the satisfactory performance and efficiency of our proposed method on large-scale real data.

Temporal action localization aims to accurately localize and recognize all possible action instances from an untrimmed video automatically. Most existing methods perform this task by first generating a set of proposals and then recognizing each independently. However, due to the complex structures and large content variations in action instances, recognizing them individually can be difficult. Fortunately, some proposals often share information regarding one specific action. Such information, which is ignored in existing methods, can be used to boost recognition performance. In this paper, we propose a novel mechanism, called relation attention, to exploit informative relations among proposals based on their appearance or optical flow features. Specifically, we propose a relation attention module to enhance representation power by capturing useful information from other proposals. This module does not change the dimensions of the original input and output and does not rely on any specific proposal generation methods or feature extraction backbone networks. Experimental results show that the proposed relation attention mechanism improves performance significantly on both Thumos14 and ActivityNet1.3 datasets compared to existing architectures. For example, relying on Structured Segment Networks (SSN), the proposed relation attention module helps to increase the mAP from 41.4 to 43.7 on the Thumos14 dataset and outperforms the state-of-the-art results.

The issue of data imbalance occurs in many real-world applications especially in medical diagnosis, where normal cases are usually much more than the abnormal cases. To alleviate this issue, one of the most important approaches is the oversampling method, which seeks to synthesize minority class samples to balance the numbers of different classes. However, existing methods barely consider global geometric information involved in the distribution of minority class samples, and thus may incur distribution mismatching between real and synthetic samples. In this paper, relying on optimal transport (Villani 2008), we propose an oversampling method by exploiting global geometric information of data to make synthetic samples follow a similar distribution to that of minority class samples. Moreover, we introduce a novel regularization based on synthetic samples and shift the distribution of minority class samples according to loss information. Experiments on toy and real-world data sets demonstrate the efficacy of our proposed method in terms of multiple metrics.

Unsupervised domain adaptation aims to learn a model of classifier for unlabeled samples on the target domain, given training data of labeled samples on the source domain. Impressive progress is made recently by learning invariant features via domain-adversarial training of deep networks. In spite of the recent progress, domain adaptation is still limited in achieving the invariance of feature distributions at a finer category level. To this end, we propose in this paper a new domain adaptation method called Domain-Symmetric Networks (SymNets). The proposed SymNet is based on a symmetric design of source and target task classifiers, based on which we also construct an additional classifier that shares with them its layer neurons. To train the SymNet, we propose a novel adversarial learning objective whose key design is based on a two-level domain confusion scheme, where the category-level confusion loss improves over the domain-level one by driving the learning of intermediate network features to be invariant at the corresponding categories of the two domains. Both domain discrimination and domain confusion are implemented based on the constructed additional classifier. Since target samples are unlabeled, we also propose a scheme of cross-domain training to help learn the target classifier. Careful ablation studies show the efficacy of our proposed method. In particular, based on commonly used base networks, our SymNets achieve the new state of the art on three benchmark domain adaptation datasets.

In this paper, we propose to train convolutional neural networks (CNNs) with both binarized weights and activations, leading to quantized models specifically for mobile devices with limited power capacity and computation resources. By assuming the same architecture to full-precision networks, previous works on quantizing CNNs seek to preserve the floating-point information using a set of discrete values, which we call value approximation. However, we take a novel ``structure approximation'' view for quantization--- it is very likely that a different architecture may be better for best performance. In particular, we propose a ``network decomposition'' strategy, named Group-Net, in which we divide the network into groups. In this way, each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches. In addition, we learn effect connections among groups to improve the representational capability. Moreover, the proposed Group-Net shows strong generalization to other tasks. For instance, we extend Group-Net for highly accurate semantic segmentation by embedding rich context into the binary structure. Experiments on both classification and semantic segmentation tasks demonstrate the superior performance of the proposed methods over various popular architectures. In particular, we outperform the previous best binary neural networks in terms of accuracy and huge computation saving.

Most state-of-the-art action localization systems process each action proposal individually, without explicitly exploiting their relations during learning. However, the relations between proposals actually play an important role in action localization, since a meaningful action always consists of multiple proposals in a video. In this paper, we propose to exploit the proposal-proposal relations using GraphConvolutional Networks (GCNs). First, we construct an action proposal graph, where each proposal is represented as a node and their relations between two proposals as an edge. Here, we use two types of relations, one for capturing the context information for each proposal and the other one for characterizing the correlations between distinct actions. Then we apply the GCNs over the graph to model the relations among different proposals and learn powerful representations for the action classification and localization. Experimental results show that our approach significantly outperforms the state-of-the-art on THUMOS14(49.1% versus 42.8%). Moreover, augmentation experiments on ActivityNet also verify the efficacy of modeling action proposal relationships.

There has been a prominent emergence of work on video prediction, aiming to extrapolate the future video frames from the past. Existing temporal-based methods are limited to certain numbers of frames. In this paper, we study video prediction from a single still image in the facial expression domain, a.k.a, facial image-to-video translation. Our main approach, dubbed AffineGAN, associates each facial image with an expression intensity and leverages an affine transformation in the latent space. AffineGAN allows users to control the number of frames to predict as well as the expression intensity for each of them. Unlike previous intensity-based methods, We derive an inverse formulation to the affine transformation, enabling automatic inference of the facial expression intensities from videos --- manual annotation is not only tedious but also ambiguous as people express in various ways and have different opinions about the intensity of a facial image. Both quantitative and qualitative results verify the superiority of AffineGAN over the state of the arts. Notably, in a Turing test with web faces, more than 50% of the facial expression videos generated by AffineGAN are considered real by the Amazon Mechanical Turk workers. This work could improve users' communication experience by enabling them to conveniently and creatively produce expression GIFs, which are popular art forms in online messaging and social networks.

Multiple marginal matching problem aims at learning mappings to match a source domain to multiple target domains and it has attracted great attention in many applications, such as multi-domain image translation. However, addressing this problem has two critical challenges: (i) Measuring the multi-marginal distance among different domains is very intractable; (ii) It is very difficult to exploit cross-domain correlations to match the target domain distributions. In this paper, we propose a novel Multi-marginal Wasserstein GAN (MWGAN) to minimize Wasserstein distance among domains. Specifically, with the help of multi-marginal optimal transport theory, we develop a new adversarial objective function with inner- and inter-domain constraints to exploit cross-domain correlations. Moreover, we theoretically analyze the generalization performance of MWGAN, and empirically evaluate it on the balanced and imbalanced translation tasks. Extensive experiments on toy and real-world datasets demonstrate the effectiveness of MWGAN.

Designing effective architectures is one of the key factors behind the success of deep neural networks. Existing deep architectures are either manually designed or automatically searched by some Neural Architecture Search (NAS) methods. However, even a well-searched architecture may still contain many non-significant or redundant modules or operations (e.g., convolution or pooling), which may not only incur substantial memory consumption and computation cost but also deteriorate the performance. Thus, it is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computation cost. Unfortunately, such a constrained optimization problem is NP-hard. To make the problem feasible, we cast the optimization problem into a Markov decision process (MDP) and seek to learn a Neural Architecture Transformer (NAT) to replace the redundant operations with the more computationally efficient ones (e.g., skip connection or directly removing the connection). Based on MDP, we learn NAT by exploiting reinforcement learning to obtain the optimization policies w.r.t. different architectures. To verify the effectiveness of the proposed strategies, we apply NAT on both hand-crafted architectures and NAS based architectures. Extensive experiments on two benchmark datasets, i.e., CIFAR-10 and ImageNet, demonstrate that the transformed architecture by NAT significantly outperforms both its original form and those architectures optimized by existing methods.

A crucial issue in heterogeneous domain adaptation (HDA) is the ability to learn a feature mapping between different types of features across domains. Inspired by language translation, a word translated from one language corresponds to only a few words in another language, we present an efficient method named Sparse Heterogeneous Feature Representation (SHFR) in this paper for multi-class HDA to learn a sparse feature transformation between domains with multiple classes. Specifically, we formulate the problem of learning the feature transformation as a compressed sensing problem by building multiple binary classifiers in the target domain as various measurement sensors, which are decomposed from the target multi-class classification problem. We show that the estimation error of the learned transformation decreases with the increasing number of binary classifiers. In other words, for adaptation across heterogeneous domains to be successful, it is necessary to construct a sufficient number of incoherent binary classifiers from the original multi-class classification problem. To achieve this, we propose to apply the error correcting output correcting (ECOC) scheme to generate incoherent classifiers. To speed up the learning of the feature transformation across domains, we apply an efficient batch-mode algorithm to solve the resultant nonnegative sparse recovery problem. Theoretically, we present a generalization error bound of our proposed HDA method under a multi-class setting. Lastly, we conduct extensive experiments on both synthetic and real-world datasets to demonstrate the superiority of our proposed method over existing state-of-the-art HDA methods in terms of prediction accuracy and training efficiency.

In many real-world applications, data can be represented by multiple ways or multi-view features to describe various characteristics of data. In this sense, the prediction performance can be significantly improved by taking advantages of these features together. Late fusion, which combines the predictions of multiple features, is a commonly used approach to make the final decision for a test instance. However, it is ubiquitous that different features dispute the prediction on the same data with each other, leading to performance degeneration. In this paper, we propose an efficient and effective matrix factorization-based approach to fuse predictions from multiple sources. This approach leverages a hard constraint on the matrix rank to preserve the consistency of predictions by various features, and we thus named it as Hard-rank Constraint Matrix Factorization-based fusion (HCMF). HCMF can avoid the performance degeneration caused by the controversy of multiple features. Extensive experiments demonstrate the efficacy of HCMF for outlier detection and the performance improvement, which outperforms the state-of-the-art late fusion algorithms on many data sets.

Total variation (TV) regularization has proven effective for a range of computer vision tasks through its preferential weighting of sharp image edges. Existing TV-based methods, however, often suffer from the over-smoothing issue and solution bias caused by the homogeneous penalization. In this paper, we consider addressing these issues by applying inhomogeneous regularization on different image components. We formulate the inhomogeneous TV minimization problem as a convex quadratic constrained linear programming problem. Relying on this new model, we propose a matching pursuit-based total variation minimization method (MPTV), specifically for image deconvolution. The proposed MPTV method is essentially a cutting-plane method that iteratively activates a subset of nonzero image gradients and then solves a subproblem focusing on those activated gradients only. Compared with existing methods, the MPTV is less sensitive to the choice of the trade-off parameter between data fitting and regularization. Moreover, the inhomogeneity of MPTV alleviates the over-smoothing and ringing artifacts and improves the robustness to errors in blur kernel. Extensive experiments on different tasks demonstrate the superiority of the proposed method over the current state of the art.

Many machine learning problems concern with discovering or associating common patterns in data of multiple views or modalities. Multi-view learning is of the methods to achieve such goals. Recent methods propose deep multi-view networks via adaptation of generic Deep Neural Networks (DNNs), which concatenate features of individual views at intermediate network layers (i.e., fusion layers). In this work, we study the problem of multi-view learning in such end-to-end networks. We take a regularization approach via multi-view learning criteria, and propose a novel, effective, and efficient neuron-wise correlation-maximizing regularizer. We implement our proposed regularizers collectively as a correlation-regularized network layer (CorrReg). CorrReg can be applied to either fully-connected or convolutional fusion layers, simply by replacing them with their CorrReg counterparts. By partitioning neurons of a hidden layer in generic DNNs into multiple subsets, we also consider a multi-view feature learning perspective of generic DNNs. Such a perspective enables us to study deep multi-view learning in the context of regularized network training, for which we present control experiments of benchmark image classification to show the efficacy of our proposed CorrReg. To investigate how CorrReg is useful for practical multi-view learning problems, we conduct experiments of RGB-D object/scene recognition and multi-view based 3D object recognition, using networks with fusion layers that concatenate intermediate features of individual modalities or views for subsequent classification. Applying CorrReg to fusion layers of these networks consistently improves classification performance. In particular, we achieve the new state of the art on the benchmark RGB-D object and RGB-D scene datasets. We make the implementation of CorrReg publicly available.

We address the challenging problem of weakly supervised temporal action localization from unconstrained web videos, where only the video-level action labels are available during training. Inspired by the adversarial erasing strategy in weakly supervised semantic segmentation, we propose a novel iterative-winners-out network. Specifically, we make two technical contributions: we propose an iterative training strategy, namely, winners-out, to select the most discriminative action instances in each training iteration and remove them in the next training iteration. This iterative process alleviates the "winner-takes-all" phenomenon that existing approaches tend to choose the video segments that strongly correspond to the video label but neglects other less discriminative video segments. With this strategy, our network is able to localize not only the most discriminative instances but also the less discriminative ones. To better select the target action instances in winners-out, we devise a class-discriminative localization technique. By employing the attention mechanism and the information learned from data, our technique is able to identify the most discriminative action instances effectively. The two key components are integrated into an end-to-end network to localize actions without using the frame-level annotations. Extensive experimental results demonstrate that our method outperforms the state-of-the-art weakly supervised approaches on ActivityNet1.3 and improves mAP from 16.9% to 20.5% on THUMOS14. Notably, even with weak video-level supervision, our method attains comparable accuracy to those employing frame-level supervisions.

Cost-Sensitive Online Classification has drawn extensive attention in recent years, where the main approach is to directly online optimize two well-known cost-sensitive metrics: (i) weighted sum of sensitivity and specificity and (ii) weighted misclassification cost. However, previous existing methods only considered first-order information of data stream. It is insufficient in practice, since many recent studies have proved that incorporating second-order information enhances the prediction performance of classification models. Thus, we propose a family of cost-sensitive online classification algorithms with adaptive regularization in this paper. We theoretically analyze the proposed algorithms and empirically validate their effectiveness and properties in extensive experiments. Then, for better trade off between the performance and efficiency, we further introduce the sketching technique into our algorithms, which significantly accelerates the computational speed with quite slight performance loss. Finally, we apply our algorithms to tackle several online anomaly detection tasks from real world. Promising results prove that the proposed algorithms are effective and efficient in solving cost-sensitive online classification problems in various real-world domains.

Generating images via a generative adversarial network (GAN) has attracted much attention recently. However, most of the existing GAN-based methods can only produce lowresolution images of limited quality. Directly generating highresolution images using GANs is nontrivial, and often produces problematic images with incomplete objects. To address this issue, we develop a novel GAN called auto-embedding generative adversarial network, which simultaneously encodes the global structure features and captures the fine-grained details. In our network, we use an autoencoder to learn the intrinsic high-level structure of real images and design a novel denoiser network to provide photo-realistic details for the generated images. In the experiments, we are able to produce 512 × 512 images of promising quality directly from the input noise. The resultant images exhibit better perceptual photo-realism, that is, with sharper structure and richer details, than other baselines on several datasets, including Oxford-102 Flowers, Caltech-UCSD Birds (CUB), High-Quality Large-scale CelebFaces Attributes (CelebAHQ), Large-scale Scene Understanding (LSUN), and ImageNet.

Real images often have multiple labels, i.e., each image is associated with multiple objects or attributes. Compared to single-label image classification, the multilabel classification problem is much more challenging due to several issues. At first, multiple objects can be anywhere in the image. Second, the importance of different regions in an image is different, and the regions of interest in a multilabel image can be very different from another one. Finally, multiple labels of an image can have label dependencies due to complex image structures. To address these challenges, in this paper, we propose to predict the labels sequentially by applying the recurrent neural networks (RNNs), which are used to encode the label dependencies. When predicting a specific label, we introduce a dynamic attention mechanism to enable the model to focus on only regions of interest in the image. Two benchmark datasets (i.e., Pascal VOC and MS-COCO) are adopted to demonstrate the effectiveness of our work. Moreover, we construct a new dataset, which includes many semantic dependent labels in each image, to verify the effectiveness of our model. Experimental results show that our method outperforms several state-of-the-arts, especially when predicting some semantic relative labels.

Batch Normalization (BN) has been a standard component in designing deep neural networks (DNNs). Although the standard BN can significantly accelerate the training of DNNs and improve the generalization performance, it has several underlying limitations which may hamper the performance in both training and inference. In the training stage, BN relies on estimating the mean and variance of data using a single mini-batch. Consequently, BN can be unstable when the batch size is very small or the data is poorly sampled. In the inference stage, BN often uses the so called moving mean and moving variance instead of batch statistics, i.e., the training and inference rules in BN are not consistent. Regarding these issues, we propose a memorized batch normalization (MBN), which considers multiple recent batches to obtain more accurate and robust statistics. Note that after the SGD update for each batch, the model parameters will change, and the features will change accordingly, leading to the Distribution Shift before and after the update for the considered batch. To alleviate this issue, we present a simple Double-Forward scheme in MBN which can further improve the performance. Compared to related methods, the proposed MBN exhibits consistent behaviors in both training and inference. Empirical results show that the MBN based models trained with the Double-Forward scheme greatly reduce the sensitivity of data and significantly improve the generalization performance.

Video-based action recognition with deep neural networks has shown remarkable progress. However, most of the existing approaches are too computationally expensive due to the complex network architecture. To address these problems, we propose a new real-time action recognition architecture, called Temporal Convolutional 3D Network (T-C3D), which learns video action representations in a hierarchical multi-granularity manner. Specifically, we combine a residual 3D convolutional neural network which captures complementary information on the appearance of a single frame and the motion between consecutive frames with a new temporal encoding method to explore the temporal dynamics of the whole video. Thus heavy calculations are avoided when doing the inference, which enables the method to be capable of real-time processing. On two challenging benchmark datasets, UCF101 and HMDB51, our method is significantly better than state-of-the-art real-time methods by over 5.4% in terms of accuracy and 2 times faster in terms of inference speed (969 frames per second), demonstrating comparable recognition performance to the state-of-the-art methods. The source code for the complete system as well as the pre-trained models are publicly available at https://github.com/tc3d.

Cartoon-to-photo facial translation could be widely used in different applications, such as law enforcement and anime remaking. Nevertheless, current general-purpose image-to-image models \ygyan{usually} %can only produce blurry or unrelated results in this task. In this paper, we propose a Cartoon-to-Photo facial translation with Generative Adversarial Networks (\name) for inverting cartoon faces to generate photo-realistic and related face images. In order to produce convincing faces with intact facial parts, we exploit global and local discriminators to capture global facial features and three local facial regions, respectively. Moreover, we use a specific content network to capture and preserve face characteristic and identity between cartoons and photos. As a result, the proposed approach can generate convincing high-quality faces that satisfy both the characteristic and identity constraints of input cartoon faces. Compared with recent works on unpaired image-to-image translation, our proposed method is able to generate more realistic and correlative images.

Multi-tasking optimization has now become a promising research topic that has attracted increasing attention from researchers. In this paper, an efficient memetic evolutionary multi-tasking optimization framework is proposed. The key idea is to use multiple subpopulations to solve multiple tasks, with each subpopulation focusing on solving a single task. A knowledge transferring crossover is proposed to transfer knowledge between subpopulations during the evolution. The proposed framework is further integrated with a multi-objective differential evolution and an adaptive local search strategy, forming a memetic multiobjective DE named MM-DE for multi-tasking optimization. The proposed MM-DE is compared with the state-of-the-art multi-tasking multi-objective evolutionary algorithm (named MO-MFEA) on nine benchmark problems in the CEC 2017 multi-tasking optimization competition. The experimental results have demonstrated that the proposed MM-DE can offer very promising performance.

Visual Grounding (VG) aims to locate the most relevant object or region in an image, based on a natural language query. The query can be a phrase, a sentence or even a multi-round dialogue. There are three main challenges in VG: 1) what is the main focus in a query; 2) how to understand an image; 3) how to locate an object. Most existing methods combine all the information curtly, which may suffer from the problem of information redundancy (i.e. ambiguous query, complicated image and a large number of objects). In this paper, we formulate these challenges as three attention problems and propose an accumulated attention (A-ATT) mechanism to reason among them jointly. Our A-ATT mechanism can circularly accumulate the attention for useful information in image, query, and objects, while the noises are ignored gradually. We evaluate the performance of A-ATT on four popular datasets (namely Refer-COCO, ReferCOCO+, ReferCOCOg, and Guesswhat?!), and the experimental results show the superiority of the proposed method in term of accuracy.

This paper tackles the problem of training a deep convolutional neural network with both low-precision weights and low-bitwidth activations. Optimizing a low-precision network is very challenging since the training process can easily get trapped in a poor local minima, which results in substantial accuracy loss. To mitigate this problem, we propose three simple-yet-effective approaches to improve the network training. First, we propose to use a two-stage optimization strategy to progressively find good local minima. Specifically, we propose to first optimize a net with quantized weights and then quantized activations. This is in contrast to the traditional methods which optimize them simultaneously. Second, following a similar spirit of the first method, we propose another progressive optimization approach which progressively decreases the bit-width from high-precision to low-precision during the course of training. Third, we adopt a novel learning scheme to jointly train a full-precision model alongside the low-precision one. By doing so, the full-precision model provides hints to guide the low-precision model training. Extensive experiments on various datasets (i.e., CIFAR-100 and ImageNet) show the effectiveness of the proposed methods. To highlight, using our methods to train a 4-bit precision network leads to no performance decrease in comparison with its full-precision counterpart with standard network architectures (i.e., AlexNet and ResNet-50).

Generative adversarial networks (GANs) aim to generate realistic data from some prior distribution (e.g., Gaussian noises). However, such prior distribution is often independent of real data and thus may lose semantic information (e.g., geometric structure or content in images) of data. In practice, the semantic information might be represented by some latent distribution learned from data, which, however, is hard to be used for sampling in GANs. In this paper, rather than sampling from the pre-defined prior distribution, we propose a Local Coordinate Coding (LCC) based sampling method to improve GANs. We derive a generalization bound for LCC based GANs and prove that a small dimensional input is sufficient to achieve good generalization. Extensive experiments on various real-world datasets demonstrate the effectiveness of the proposed method.

Heterogeneous domain adaptation (HDA) aims to exploit knowledge from a heterogeneous source domain to improve the learning performance in a target domain. Since the feature spaces of the source and target domains are different, the transferring of knowledge is extremely difficult. In this paper, we propose a novel semi-supervised algorithm for HDA by exploiting the theory of optimal transport (OT), a powerful tool originally designed for aligning two different distributions. To match the samples between heterogeneous domains, we propose to preserve the semantic consistency between heterogeneous domains by incorporating label information into the entropic Gromov-Wasserstein discrepancy, which is a metric in OT for different metric spaces, resulting in a new semi-supervised scheme. Via the new scheme, the target and transported source samples with the same label are enforced to follow similar distributions. Lastly, based on the Kullback-Leibler metric, we develop an efficient algorithm to optimize the resultant problem. Comprehensive experiments on both synthetic and real-world datasets demonstrate the effectiveness of our proposed method.

Multimodel wearable sensor data classificationplays an important role in ubiquitous computingand has a wide range of applications in variousscenarios from healthcare to entertainment. How-ever, most of the existing work in this field em-ploys domain-specific approaches and is thus inef-fective in complex situations where multi-modalitysensor data is collected. Moreover, the wearablesensor data is less informative than the conven-tional data such as texts or images. In this paper,to improve the adaptability of such classificationmethods across different application contexts, weturn this classification task into a game and applya deep reinforcement learning scheme to dynami-cally deal with complex situations. We also intro-duce a selective attention mechanism into the rein-forcement learning scheme to focus on the crucialdimensions of the data. This mechanism helps tocapture extra information from the signal, and canthus significantly improve the discriminative powerof the classifier. We carry out several experimentson three wearable sensor datasets, and demonstratecompetitive performance of the proposed approachcompared to several state-of-the-art baselines.

This paper investigates Online Active Learning (OAL) for imbalanced unlabeled datastream, where only a budget of labels can be queried to optimize some cost-sensitive performance measure. OAL can solve many real-world problems, such as anomaly detection in healthcare, finance and network security. In these problems, there are two key challenges: the query budget is often limited; the ratio between two classes is highly imbalanced. To address these challenges, existing work of OAL adopts either asymmetric losses or queries (an isolated asymmetric strategy) to tackle the imbalance, and uses first-order methods to optimize the cost-sensitive measure. However, they may incur two deficiencies: (1) the poor ability in handling imbalanced data due to the isolated asymmetric strategy; (2) relative slow convergence rate due to the first-order optimization. In this paper, we propose a novel Online Adaptive Asymmetric Active (OA3) learning algorithm, which is based on a new asymmetric strategy (merging both the asymmetric losses and queries strategies), and second-order optimization. We theoretically analyze its bounds, and also empirically evaluate it on four real-world online anomaly detection tasks. Promising results confirm the effectiveness and robustness of the proposed algorithm in various application domains.

Channel pruning is one of the predominant approaches for deep model compression. Existing pruning methods either train from scratch with sparsity constraints on channels, or minimize the reconstruction error between the pre-trained feature maps and the compressed ones. Both strategies suffer from some limitations: the former kind is computationally expensive and difficult to converge, whilst the latter kind optimizes the reconstruction error but ignores the discriminative power of channels. To overcome these drawbacks, we investigate a simple-yet-effective method, called discrimination-aware channel pruning, to choose those channels that really contribute to discriminative power. To this end, we introduce additional losses into the network to increase the discriminative power of intermediate layers and then select the most discriminative channels for each layer by considering the additional loss and the reconstruction error. Last, we propose a greedy algorithm to conduct channel selection and parameter optimization in an iterative way. Extensive experiments demonstrate the effectiveness of our method. For example, on ILSVRC-12, our pruned ResNet-50 with 30% reduction of channels even outperforms the original model by 0.39% in top-1 accuracy.

Understanding and recognizing human activities is a fundamental research topic for a wide range of important applications such as fall detection and remote health monitoring and intervention. Despite active research in human activity recognition over the past years, existing approaches based on computer vision or wearable sensor technologies present several significant issues such as privacy (e.g., using video camera to monitor the elderly at home) and practicality (e.g., not possible for an older person with dementia to remember wearing devices). In this paper, we present a low-cost, unobtrusive, and robust system that supports independent living of older people. The system interprets what a person is doing by deciphering signal fluctuations using radio-frequency identification (RFID) technology and machine learning algorithms. To deal with noisy, streaming, and unstable RFID signals, we develop a compressive sensing, dictionary-based approach that can learn a set of compact and informative dictionaries of activities using an unsupervised subspace decomposition. In particular, we devise a number of approaches to explore the properties of sparse coefficients of the learned dictionaries for fully utilizing the embodied discriminative information on the activity recognition task. Our approach achieves efficient and robust activity recognition via a more compact and robust representation of activities. Extensive experiments conducted in a real-life residential environment demonstrate that our proposed system offers a good overall performance and shows the promising practical potential to underpin the applications for the independent living of the elderly.

In this paper, we study the online heterogeneous transfer (OHT) learning problem, where the target data of interest arrive in an online manner, while the source data and auxiliary co-occurrence data are from offline sources and can be easily annotated. OHT is very challenging, since the feature spaces of the source and target domains are different. To address this, we propose a novel technique called OHT by hedge ensemble by exploiting both offline knowledge and online knowledge of different domains. To this end, we build an offline decision function based on a heterogeneous similarity that is constructed using labeled source data and unlabeled auxiliary co-occurrence data. After that, an online decision function is learned from the target data. Last, we employ a hedge weighting strategy to combine the offline and online decision functions to exploit knowledge from the source and target domains of different feature spaces. We also provide a theoretical analysis regarding the mistake bounds of the proposed approach. Comprehensive experiments on three real-world data sets demonstrate the effectiveness of the proposed technique.

Unlike traditional LASSO enforcing sparsity on the variables, Generalized LASSO (GL) enforces sparsity on a linear transformation of the variables, gaining flexibility and success in many applications. However, many existing GL algorithms do not scale up to high-dimensional problems, and/or only work well for a specific choice of the transformation. We propose an efficient Matching Pursuit Generalized LASSO (MPGL) method, which overcomes these issues, and is guaranteed to converge to a global optimum. We formulate the GL problem as a convex quadratic constrained linear programming (QCLP) problem and tailor-make a cutting plane method. More specifically, our MPGL iteratively activates a subset of nonzero elements of the transformed variables, and solves a subproblem involving only the activated elements thus gaining significant speed-up. Moreover, MPGL is less sensitive to the choice of the trade-off hyper-parameter between data fitting and regularization, and mitigates the long-standing hyper-parameter tuning issue in many existing methods. Experiments demonstrate the superior efficiency and accuracy of the proposed method over the state-of-the-arts in both classification and image processing tasks.

Deep learning has achieved unprecedented practical success in many applications. Despite its empirical success, however, the theoretical understanding of deep neural networks still remains a major open problem. In this paper, we explore properties of two-layered ReLU networks. For simplicity, we assume that the optimal model parameters (also called ground-truth parameters) are known. We then assume that a network receives Gaussian input and is trained by minimizing the expected squared loss between the prediction function of the network and a target function. To conduct the analysis, we propose a normal equation for critical points, and study the invariances under three kinds of transformations, namely, scale transformation, rotation transformation and perturbation transformation. We prove that these transformations can keep the loss of a critical point invariant, thus can incur flat regions. Consequently, how to escape from flat regions is vital in training neural networks.

The challenge in blind image deblurring is to remove the effects of blur with limited prior information about the nature of the blur process. Existing methods often assume that the blur image is produced by linear convolution with additive Gaussian noise. However, including even a small number of outliers to this model in the kernel estimation process can significantly reduce the resulting image quality. Previous methods mainly rely on some simple but unreliable heuristics to identify outliers for kernel estimation. Rather than attempt to identify outliers to the model a priori, we instead propose to sequentially identify inliers, and gradually incorporate them into the estimation process. The self-paced kernel estimation scheme we propose represents a generalization of existing self-paced learning approaches, in which we gradually detect and include reliable inlier pixel sets in a blurred image for kernel estimation. Moreover, we automatically activate a subset of significant gradients w.r.t. the reliable inlier pixels, and then update the intermediate sharp image and the kernel accordingly. Experiments on both synthetic data and real-world images with various kinds of outliers demonstrate the effectiveness and robustness of the proposed method compared to the stateof- the-art methods.

In this paper, we study relations ranking and object classification for multi-relational data where objects are interconnected by multiple relations. The relations among objects should be exploited for achieving a good classification. While most existing approaches exploit either by directly counting the number of connections among objects or by learning the weight of each relation from labeled data only. In this paper, we propose an algorithm, TensorRRCC, which is able to determine the ranking of relations and the labels of objects simultaneously. Our basic idea is that highly ranked relations within a class should play more important roles in object classification, and class membership information is important for determining a ranking quality over the relations w.r.t. a specific learning task. TensorRRCC implements the idea by modeling a Markov chain on transition probability graphs from connection and feature information with both labeled and unlabeled objects and propagates the ranking scores of relations and relevant classes of objects. An iterative progress is proposed to solve a set of tensor equations to obtain the stationary distribution of relations and objects. We compared our algorithm with current collective classification algorithms on two real-world data sets and the experimental results show the superiority of our method.

Domain adaptation aims to reduce the effort on collecting and annotating target data by leveraging knowledge from a different source domain. The domain adaptation problem will become extremely challenging when the feature spaces of the source and target domains are different, which is also known as the heterogeneous domain adaptation (HDA) problem. In this paper, we propose a novel HDA method to find the optimal discriminative correlation subspace for the source and target data. The discriminative correlation subspace is inherited from the canonical correlation subspace between the source and target data, and is further optimized to maximize the discriminative ability for the target domain classifier. We formulate a joint objective in order to simultaneously learn the discriminative correlation subspace and the target domain classifier. We then apply an alternating direction method of multiplier (ADMM) algorithm to address the resulting non-convex optimization problem. Comprehensive experiments on two real-world data sets demonstrate the effectiveness of the proposed method compared to the state-of-the-art methods.

Transfer learning techniques have been broadly applied in applications where labeled data in a target domain are difficult to obtain while a lot of labeled data are available in related source domains. In practice, there can be multiple source domains that are related to the target domain, and how to combine them is still an open problem. In this paper, we seek to leverage labeled data from multiple source domains to enhance classification performance in a target domain where the target data are received in an online fashion. This problem is known as the online transfer learning problem. To achieve this, we propose novel online transfer learning paradigms in which the source and target domains are leveraged adaptively. We consider two different problem settings: homogeneous transfer learning and heterogeneous transfer learning. The proposed methods work in an online manner, where the weights of the source domains are adjusted dynamically. We provide the mistake bounds of the proposed methods and perform comprehensive experiments on real-world data sets to demonstrate the effectiveness of the proposed algorithms.

Confidence-weighted (CW) learning is a successful online learning paradigm which maintains a Gaussian distribution over classifier weights and adopts a covariancematrix to represent the uncertainties of the weight vectors. However, there are two deficiencies in existing full CW learning paradigms, these being the sensitivity to irrelevant features, and the poor scalability to high dimensional data due to the maintenance of the covariance structure. In this paper, we begin by presenting an online-batch CW learning scheme, and then present a novel paradigm to learn sparse CW classifiers. The proposed paradigm essentially identifies feature groups and naturally builds a block diagonal covariance structure, making it very suitable for CW learning over very high-dimensional data.Extensive experimental results demonstrate the superior performance of the proposed methods over state-of-the-art counterparts on classification and feature selection tasks.

Orthogonal Non-negative Matrix Factorization (ONMF) approximates a data matrix X by the product of two lower dimensional factor matrices: X -- UVT, with one of them orthogonal. ONMF has been widely applied for clustering, but it often suffers from high computational cost due to the orthogonality constraint. In this paper, we propose a method, called Nonlinear Riemannian Conjugate Gradient ONMF (NRCG-ONMF), which updates U and V alternatively and preserves the orthogonality of U while achieving fast convergence speed. Specifically, in order to update U, we develop a Nonlinear Riemannian Conjugate Gradient (NRCG) method on the Stiefel manifold using Barzilai-Borwein (BB) step size. For updating V, we use a closed-form solution under non-negativity constraint. Extensive experiments on both synthetic and real-world data sets show consistent superiority of our method over other approaches in terms of orthogonality preservation, convergence speed and clustering performance.

Blind image deconvolution is an ill-posed inverse problem which is often addressed through the application of appropriate prior. Although some priors are informative in general, many images do not strictly conform to this, leading to degraded performance in the kernel estimation. More critically, real images may be contaminated by nonuniform noise such as saturation and outliers. Methods for removing specific image areas based on some priors have been proposed, but they operate either manually or by defining fixed criteria. We show here that a subset of the image gradients are adequate to estimate the blur kernel robustly, no matter the gradient image is sparse or not. We thus introduce a gradient activation method to automatically select a subset of gradients of the latent image in a cutting-plane-based optimization scheme for kernel estimation. No extra assumption is used in our model, which greatly improves the accuracy and flexibility. More importantly, the proposed method affords great convenience for handling noise and outliers. Experiments on both synthetic data and real-world images demonstrate the effectiveness and robustness of the proposed method in comparison with the state-of-the-art methods.

The SVM+ approach has shown excellent performance in visual recognition tasks for exploiting privileged information in the training data. In this paper, we propose two efficient algorithms for solving the linear and kernel SVM+, respectively. For linear SVM+, we absorb the bias term into the weight vector, and formulate a new optimization problem with simpler constraints in the dual form. Then, we develop an efficient dual coordinate descent algorithm to solve the new optimization problem. For kernel SVM+, we further apply the l2-loss, which leads to a simpler optimization problem in the dual form with only half of dual variables when compared with the dual form of the original SVM+ method. More interestingly, we show that our new dual problem can be efficiently solved by using the SMO algorithm of the one-class SVM problem. Comprehensive experiments on three datasets clearly demonstrate that our proposed algorithms achieve significant speed-up than the state-of-the-art solvers for linear and kernel SVM+.

Trace-norm regularization plays an important role in many areas such as computer vision and machine learning. When solving general large-scale trace-norm regularized problems, existing methods may be computationally expensive due to many high-dimensional truncated singular value decompositions (SVDs) or the unawareness of matrix ranks. In this paper, we propose a proximal Riemannian pursuit (PRP) paradigm which addresses a sequence of trace-norm regularized subproblems defined on nonlinear matrix varieties. To address the subproblem, we extend the proximal gradient method on vector space to nonlinear matrix varieties, in which the SVDs of intermediate solutions are maintained by cheap low-rank QR decompositions, therefore making the proposed method more scalable. Empirical studies on several tasks, such as matrix completion and low-rank representation based subspace clustering, demonstrate the competitive performance of the proposed paradigms over existing methods.

Multi-label classification deals with the problem where each example is associated with multiple class labels. Since the labels are often dependent to other labels, exploiting label dependencies can significantly improve the multi-label classification performance. The label dependency in existing studies is often given as prior knowledge or learned from the labels only. However, in many real applications, such prior knowledge may not be available, or labeled information might be very limited. In this paper, we propose a new algorithm, called Ml-Forest , to learn an ensemble of hierarchical multi-label classifier trees to reveal the intrinsic label dependencies. In Ml-Forest, we construct a set of hierarchical trees, and develop a label transfer mechanism to identify the multiple relevant labels in a hierarchical way. In general, the relevant labels at higher levels of the trees capture more discriminable label concepts, and they will be transferred into lower level children nodes that are harder to discriminate. The relevant labels in the hierarchy are then aggregated to compute label dependency and make the final prediction. Our empirical study shows encouraging results of the proposed algorithm in comparison with the state-of-the-art multi-label classification algorithms under Friedman test and post-hoc Nemenyi test.

Recently, low-rank representation (LRR) has shown promising performance in many real-world applications such as face clustering. However, LRR may not achieve satisfactory results when dealing with the data from nonlinear subspaces, since it is originally designed to handle the data from linear subspaces in the input space. Meanwhile, the kernel-based methods deal with the nonlinear data by mapping it from the original input space to a new feature space through a kernel-induced mapping. To effectively cope with the nonlinear data, we first propose the kernelized version of LRR in the clean data case. We also present a closed-form solution for the resultant optimization problem. Moreover, to handle corrupted data, we propose the robust kernel LRR (RKLRR) approach, and develop an efficient optimization algorithm to solve it based on the alternating direction method. In particular, we show that both the subproblems in our optimization algorithm can be efficiently and exactly solved, and it is guaranteed to obtain a globally optimal solution. Besides, our proposed algorithm can also solve the original LRR problem, which is a special case of our RKLRR when using the linear kernel. In addition, based on our new optimization technique, the kernelization of some variants of LRR can be similarly achieved. Comprehensive experiments on synthetic data sets and real-world data sets clearly demonstrate the efficiency of our algorithm, as well as the effectiveness of RKLRR and the kernelization of two variants of LRR.

Exploiting label dependency for multi-label image classification can significantly improve classification performance. Probabilistic Graphical Models are one of the primary methods for representing such dependencies. The structure of graphical models, however, is either determined heuristically or learned from very limited information. Moreover, neither of these approaches scales well to large or complex graphs. We propose a principled way to learn the structure of a graphical model by considering input features and labels, together with loss functions. We formulate this problem into a max-margin framework initially, and then transform it into a convex programming problem. Finally, we propose a highly scalable procedure that activates a set of cliques iteratively. Our approach exhibits both strong theoretical properties and a significant performance improvement over state-of-the-art methods on both synthetic and real-world data sets.

Collective classification (CC) is a task to jointly classifying related instances of network data. Enabling CC usually improves the performance of predictive models on fully-labeled training networks with large amount of labeled data. However, acquiring such labels can be difficult and costly, and learning a CC classifier with only a few labeled data can lead to poor performance. On the other hand, there are usually large amount of unlabeled data available in practical. This naturally motivates semi-supervised collective classification (SSCC) approaches for leveraging the unlabeled data to improve CC from a sparsely-labeled network. In this paper, we propose a novel non-negative matrix factorization (NMF) based SSCC algorithm, called NMF-SSCC, to effectively learn a data representation by exploiting both labeled and unlabeled data on the network. Our idea is to use matrix factorization to obtain a compact representation of network data which uncovers the class discrimination of the data inferred from the labeled instances and simultaneously respects the intrinsic network structure. To achieve this, we design a new matrix factorization objective function and incorporate a label matrix factorization term as well as a network regularization term into it. An efficient optimization algorithm using the multiplicative updating rules is then developed to solve the new objective function. To further boost the predicting performance, we extend the proposed NMF-SSCC method into an ensemble scheme, called NMFE-SSCC, in terms of building a classification ensemble with a set of NMF-SSCC collective classifiers using different constructed latent graphs. Each NMF-SSCC classifier is learnt from one latent graph generated with various latent linkages for effectively label propagation. Experimental results on real-world data sets have demonstrated the effectiveness of the new methods.

Although graph embedding has been a powerful tool for modeling data intrinsic structures, simply employing all features for data structure discovery may result in noise amplification. This is particularly severe for high dimensional data with small samples. To meet this challenge, this paper proposes a novel efficient framework to perform feature selection for graph embedding, in which a category of graph embedding methods is cast as a least squares regression problem. In this framework, a binary feature selector is introduced to naturally handle the feature cardinality in the least squares formulation. The resultant integral programming problem is then relaxed into a convex Quadratically Constrained Quadratic Program (QCQP) learning problem, which can be efficiently solved via a sequence of accelerated proximal gradient (APG) methods. Since each APG optimization is w.r.t. only a subset of features, the proposed method is fast and memory efficient. The proposed framework is applied to several graph embedding learning problems, including supervised, unsupervised, and semi-supervised graph embedding. Experimental results on several high dimensional data demonstrated that the proposed method outperformed the considered state-of-the-art methods.

In this paper, we present an efficient Multi-class Heterogeneous Domain Adaptation (HDA) method, where data from the source and target domains are represented by heterogeneous features with different dimensions. Specifically, we propose to reconstruct a sparse feature transformation matrix to map the features of multiple classes from the source domain to the target domain. We cast this learning task as a compressed sensing problem, where each classifier can be deemed as a measurement sensor. Based on compressive sensing theory, the estimation error of the transformation matrix decreases with the increasing number of classifiers. Therefore, to guarantee the reconstruction performance, we construct sufficiently many binary classifiers based on the error correcting output correcting. Extensive experiments are conducted on both toy data and three real-world HDA applications to verify the superiority of our proposed method over existing state-of-the-art HDA methods in terms of prediction accuracy.

Low rank matrix recovery is a fundamental task in many real-world applications. The performance of existing methods, however, deteriorates significantly when applied to ill-conditioned or large-scale matrices. In this paper, we therefore propose an efficient method, called Riemannian Pursuit (RP), that aims to address these two problems simultaneously. Our method consists of a sequence of fixed-rank optimization problems. Each subproblem, solved by a nonlinear Riemannian conjugate gradient method, aims to correct the solution in the most important subspace of increasing size. Theoretically, RP converges linearly under mild conditions and experimental results show that it substantially outperforms existing methods when applied to large-scale and ill-conditioned matrices.

Because of the strong convexity and probabilistic underpinnings, logistic regression (LR) is widely used in many real-world applications. However, in many problems, such as bioinformatics, choosing a small subset of features with the most discriminative power are desirable for interpreting the prediction model, robust predictions or deeper analysis. To achieve a sparse solution with respect to input features, many sparse LR models are proposed. However, it is still challenging for them to efficiently obtain unbiased sparse solutions to very high-dimensional problems (e.g., identifying the most discriminative subset from millions of features). In this paper, we propose a new minimax sparse LR model for very high-dimensional feature selections, which can be efficiently solved by a cutting plane algorithm. To solve the resultant nonsmooth minimax subproblems, a smoothing coordinate descent method is presented. Numerical issues and convergence rate of this method are carefully studied. Experimental results on several synthetic and real-world datasets show that the proposed method can obtain better prediction accuracy with the same number of selected features and has better or competitive scalability on very high-dimensional problems compared with the baseline methods, including the l1-regularized LR.

In this paper, a new convex matching pursuit scheme is proposed for tackling large-scale sparse coding and subset selection problems. In contrast with current matching pursuit algorithms such as subspace pursuit (SP), the proposed algorithm has a convex formulation and guarantees that the objective value can be monotonically decreased. Moreover, theoretical analysis and experimental results show that the proposed method achieves better scalability while maintaining similar or better decoding ability compared with state-of-the-art methods on large-scale problems.

Particle swarm optimizer (PSO) is a powerful optimization algorithm that has been applied to a variety of problems. It can, however, suffer from premature convergence and slow convergence rate. Motivated by these two problems, a hybrid global optimization strategy combining PSOs with a modified Broyden-Fletcher-Goldfarb-Shanno (BFGS) method is presented in this paper. The modified BFGS method is integrated into the context of the PSOs to improve the particles' local search ability. In addition, in conjunction with the territory technique, a reposition technique to maintain the diversity of particles is proposed to improve the global search ability of PSOs. One advantage of the hybrid strategy is that it can effectively find multiple local solutions or global solutions to the multimodal functions in a box-constrained space. Based on these local solutions, a reconstruction technique can be adopted to further estimate better solutions. The proposed method is compared with several recently developed optimization algorithms on a set of 20 standard benchmark problems. Experimental results demonstrate that the proposed approach can obtain high-quality solutions on multimodal function optimization problems.