Generalist Virtual Agents (GVAs) powered by Multimodal Large Language Models (MLLMs) exhibit impressive capabilities. However, their long-term learning is hampered by a core limitation: a failure to evolve beyond existing trajectories. This stems from memory systems that treat experiences as isolated fragments and rely on brittle semantic retrieval, preventing the synthesis of novel solutions from disparate knowledge. To address this, we introduce CA3Mem, a framework inspired by the human hippocampus that organizes experiences into a structured memory graph. Leveraging this graph, CA3Mem features two key innovations: 1) a generative memory recombination mechanism that synthesizes novel solutions to drive agent evolution, and 2) an associative retrieval algorithm that employs spreading activation to recall a comprehensive and contextually-aware set of experiences. Experiments on OSWorld and WebArena demonstrate that CA3Mem significantly enhances agent capabilities, leading to marked improvements in long-horizon planning, compositional generalization for novel tasks, and continuous adaptation from experience.

Jie Cao, Tianwei Lin, Bo Yuan, Rolan Yan, Hongyang He, Wenqiao Zhang, Juncheng Li, Dongping Zhang, Siliang Tang, Yueting Zhuang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Haoyu Zheng, Yun Zhu, Yuqian Yuan, Bo Yuan, Wenqiao Zhang, Siliang Tang, Jun Xiao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Knowledge editing enables multimodal large language models (MLLMs) to efficiently update outdated or incorrect information. However, existing benchmarks primarily emphasize cognitive-level modifications while lacking a focus on deeper meta-cognitive processes. To bridge this gap, we introduce CogEdit, a novel benchmark designed to evaluate MLLMs' meta-cognitive knowledge editing abilities across three levels: (1) Counterfactual-Driven Editing, assessing self-awareness of knowledge correctness changes; (2) Boundary Constraint Editing, ensuring appropriate generalization without unintended interference; and (3) Noise-Robust Editing, promoting reflective evaluation of uncertain information. To advance meta-cognitive editing, we propose MIND (Meta-cognitive INtegrated Dynamic Knowledge Editing), a framework that constructs a meta-knowledge memory for self-awareness, employs game-theoretic interactions to monitor knowledge activation, and incorporates label refinement for noise-robust updates. Extensive experiments show that MIND significantly outperforms existing cognitive editing approaches, achieving strong performance on both traditional and meta-cognitive knowledge editing benchmarks.

Large Language Models (LLMs) exhibit remarkable proficiency in understanding and managing text-based tasks.Many works try to transfer these capabilities to the video domain, which are referred to as Video-LLMs. However, current Video-LLMs can only grasp the coarse-grained semantics and are unable to efficiently handle tasks involving the comprehension or localization of specific video segments. To address these challenges, we propose Momentor, a Video-LLM designed to perform fine-grained temporal understanding tasks. To facilitate the training of Momentor, we develop an automatic data generation engine to build Moment-10M, a large-scale video instruction dataset with segment-level instruction data. Building upon the foundation of the previously published Momentor and the Moment-10M dataset, we further extend this work by introducing a Spatio-Temporal Token Consolidation (STTC) method, which can merge redundant visual tokens spatio-temporally in a parameter-free manner, thereby significantly promoting computational efficiency while preserving fine-grained visual details. We integrate STTC with Momentor to develop Momentor++ and validate its performance on various benchmarks. Momentor demonstrates robust capabilities in fine-grained temporal understanding and localization. Further, Momentor++ excels in efficiently processing and analyzing extended videos with complex events, showcasing marked advancements in handling extensive temporal contexts.

Prompt tuning, a recently emerging paradigm, adapts vision-language pre-trained models to new tasks efficiently by learning "soft prompts" for frozen models. However, in few-shot scenarios, its effectiveness is limited by sensitivity to the initialization and the time-consuming search for optimal initialization, hindering rapid adaptation. Additionally, prompt tuning risks reducing the models' generalizability due to overfitting on scarce training samples. To overcome these challenges, we introduce a novel Gradient-RegulAted Meta-prompt learning (GRAM) framework that jointly meta-learns an efficient soft prompt initialization for better adaptation and a lightweight gradient regulating function for strong cross-domain generalizability in a meta-learning paradigm using only the weakly labeled image-text pre-training data. This is achieved through a Cross-Modal Hierarchical Clustering algorithm that organizes extensive image-text data into a structured hierarchy, facilitating robust meta-learning across diverse domains. Rather than designing a specific prompt tuning method, our GRAM can be easily incorporated into various prompt tuning methods in a model-agnostic way and bring about consistent improvement for them. Further, we consider a more practical but challenging setting: test-time prompt tuning with only unlabeled test samples and propose an improved structure-induced gradient regulating function to leverage the structured semantics of the meta-learning data for zero-shot generalization. This novel approach exploits the hierarchically clustered meta-learning data to model relationships between test-time data and meta-learning prototypes, facilitating the transfer of invariant knowledge without explicit annotations. Meanwhile, we introduce a structure complexity-informed strategy for adaptively constructing meta-training tasks and generating prototypes, which fully considers the diverse semantics within hierarchical clusters of different complexities. Comprehensive experiments demonstrate the state-of-the-art few- and zero-shot generalizability of our method.

Recently, Retrieval-Augmented Generation (RAG) has achieved remarkable success in addressing the challenges of Large Language Models (LLMs) without necessitating retraining. By referencing an external knowledge base, RAG refines LLM outputs, effectively mitigating issues such as “hallucination,” lack of domain-specific knowledge, and outdated information. However, the complex structure of relationships among different entities in databases presents challenges for RAG systems. In response, GraphRAG leverages structural information across entities to enable more precise and comprehensive retrieval, capturing relational knowledge and facilitating more accurate, context-aware responses. Given the novelty and potential of GraphRAG, a systematic review of current technologies is imperative. This article provides the first comprehensive overview of GraphRAG methodologies. We formalize the GraphRAG workflow, encompassing Graph-Based Indexing, Graph-Guided Retrieval, and Graph-Enhanced Generation. We then outline the core technologies and training methods at each stage. Additionally, we examine downstream tasks, application domains, evaluation methodologies, and industrial use cases of GraphRAG. Finally, we explore future research directions to inspire further inquiries and advance progress in the field. In order to track recent progress, we set up a repository at https://github.com/pengboci/GraphRAG-Survey .

While Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA) effectively address resource constraints during fine-tuning, their performance often falls short, especially in multidimensional task scenarios. To address this issue, one straightforward solution is to introduce task-specific LoRA as domain experts, leveraging the modeling of multiple capabilities of experts and thus enhancing the general capability of multi-task learning.Although promising, these additional components often add complexity to the training and inference process, contravening the efficiency that PEFT is designed to deliver. Considering this, we introduce an innovative PEFT method, TeamLoRA, consisting of a collaboration and competition module for LoRA experts, thus achieving the right balance of effectiveness and efficiency:(i) For collaboration, we introduce a novel knowledge sharing and organization mechanism designed to optimize hierarchical learning while enhancing the efficiency of model training and inference.(ii) For competition, we propose leveraging a game-theoretic interaction mechanism for experts, encouraging experts to transfer their domain-specific knowledge while facing diverse downstream tasks, thus enhancing the performance.By doing so, TeamLoRA elegantly connects the experts as a “Team” with internal collaboration and competition, enabling a faster and more accurate PEFT paradigm. Meanwhile, we curate a Comprehensive Multi-Task Evaluation (CME) benchmark to thoroughly assess the capability of multi-task learning. Experiments conducted on our CME and other benchmarks indicate the effectiveness and efficiency of TeamLoRA. Our project is available at https://github.com/DCDmllm/TeamLoRA.

Yaoke Wang, Yun Zhu, XintongBao XintongBao, Wenqiao Zhang, Suyang Dai, Kehan Chen, Wenqiang Li, Gang Huang, Siliang Tang, Yueting Zhuang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The decoding strategies widely used in large language models (LLMs) today are Top-$p$ Sampling and Top-$k$ Sampling, both of which are methods situated between greedy decoding and random sampling. Inspired by the concept of loss aversion from prospect theory in behavioral economics, and the endowment effect as highlighted by Richard H. Thaler, the 2017 Nobel Memorial Prize in Economic Sciences — particularly the principle that "the negative utility of an equivalent loss is approximately twice the positive utility of a comparable gain" — we have developed a new decoding strategy called Loss Sampling. We have demonstrated the effectiveness and validity of our method on several LLMs, including Llama-2, Llama-3 and Mistral. Our approach improves text quality by 4-30% across four pure text tasks while maintaining diversity in text generation. Furthermore, we also extend our method to multimodal large models (LMs) and Beam Search, demonstrating the effectiveness and versatility of Loss Sampling with improvements ranging from 1-10%.

Inspired by human cognitive behavior, we introduce visual modality to enhance the performance of pure text-based question-answering tasks with the development of multimodal models. However, obtaining corresponding images through manual annotation often entails high costs. Faced with this challenge, an intuitive strategy is to use search engines or use web scraping techniques to automatically obtain relevant image information. However, the images obtained by this strategy may be of low quality and may not match the context of the original task, which could fail to improve or even decrease performance on downstream tasks. In this paper, we propose a novel framework named “ITERATE”, aimed at retrieving and optimizing the quality of images to improve the alignment between text and images. Inspired by evolutionary algorithms in reinforcement learning and driven by the synergy of large language models (LLMs) and multimodal models, ITERATE employs a series of strategic actions such as filtering, optimizing, and retrieving to acquire higher quality images, and repeats this process over multiple generations to enhance the quality of the entire image cluster. Our experimental results on the ScienceQA, ARC-Easy, and OpenDataEval datasets also verify the effectiveness of our method, showing improvements of 3.5%, 5%, and 7%, respectively.

Recent endeavors in Multimodal Large Language Models (MLLMs) aim to unify visual comprehension and generation by combining LLM and diffusion models, the state-of-the-art in each task, respectively. Existing approaches rely on spatial visual tokens, where image patches are encoded and arranged according to a spatial order (e.g., raster scan). However, we show that spatial tokens lack the recursive structure inherent to languages, hence form an impossible language for LLM to master. In this paper, we build a proper visual language by leveraging diffusion timesteps to learn discrete, recursive visual tokens. Our proposed tokens recursively compensate for the progressive attribute loss in noisy images as timesteps increase, enabling the diffusion model to reconstruct the original image at any timestep. This approach allows us to effectively integrate the strengths of LLMs in autoregressive reasoning and diffusion models in precise image generation, achieving seamless multimodal comprehension and generation within a unified framework. Extensive experiments show that we achieve a new SOTA for multimodal comprehension and generation simultaneously compared with other MLLMs. Project Page: https://DDT-LLaMA.github.io/.

Video Large Language Models (Video-LLMs) have recently shown strong performance in basic video understanding tasks, such as captioning and coarse-grained question answering, but struggle with compositional reasoning that requires multi-step spatio-temporal inference across object relations, interactions, and events. The hurdles to enhancing this capability include extensive manual labor, the lack of spatio-temporal compositionality in existing training data and the absence of explicit reasoning supervision. In this paper, we propose STEP, a novel graph-guided self-training method that enables Video-LLMs to generate reasoning-rich fine-tuning data from any raw videos to improve itself. Specifically, we first induce Spatio-Temporal Scene Graph (STSG) representation of diverse videos to capture fine-grained, multi-granular video semantics. Then, the STSGs guide the derivation of multi-step reasoning Question-Answer (QA) data with Chain-of-Thought (CoT) rationales. Both answers and rationales are integrated as training objective, aiming to enhance model’s reasoning abilities by supervision over explicit reasoning steps. Experimental results demonstrate the effectiveness of STEP across models of varying scales, with a significant 21.3% improvement in tasks requiring three or more reasoning steps. Furthermore, it achieves superior performance with a minimal amount of self-generated rationale-enriched training samples in both compositional reasoning and comprehensive understanding benchmarks, highlighting the broad applicability and vast potential.

Instruction-based image editing aims to modify specific image elements with natural language instructions. However, current models in this domain often struggle to execute complex user instructions accurately, as they are trained on low-quality data with limited editing types. We present AnyEdit, a comprehensive multi-modal instruction editing dataset, comprising 2.5 million high-quality editing pairs spanning over 20 editing types and five domains. We ensure the diversity and quality of the AnyEdit collection through three aspects: initial data diversity, adaptive editing process, and automated selection of editing results. Using the dataset, we further train a novel AnyEdit Stable Diffusion with task-aware routing and learnable task embedding for unified image editing. Comprehensive experiments on three benchmark datasets show that AnyEdit consistently boosts the performance of diffusion-based editing models. This presents prospects for developing instruction-driven image editing models that support human creativity.

Recent advancements in reward signal usage for Large Language Models (LLMs) are remarkable. However, significant challenges exist when transitioning reward signal to the multimodal domain, including labor-intensive annotations, over-reliance on one-step rewards, and inadequate evaluation. To address these issues, we propose SVIP, a novel approach to train a step-level multi-dimensional Chain-of-Thought (CoT) reward model automatically. It generates code for solving visual tasks and transforms the analysis of code blocks into the evaluation of CoT step as training samples. Then, we train SVIP-Reward model using a multi-head attention mechanism called TriAtt-CoT. The advantages of SVIP-Reward are evident throughout the entire process of MLLM. We also introduce a benchmark for CoT reward model training and testing. Experimental results demonstrate that SVIP-Reward improves MLLM performance across training and inference-time scaling, yielding better results on benchmarks while reducing hallucinations and enhancing reasoning ability.

Digital agents are increasingly employed to automate tasks in interactive digital environments such as web pages, software applications, and operating systems. While text-based agents built on Large Language Models (LLMs) often require frequent updates due to platform-specific APIs, visual agents leveraging Multimodal Large Language Models (MLLMs) offer enhanced adaptability by interacting directly with Graphical User Interfaces (GUIs). However, these agents face significant challenges in visual perception, particularly when handling high-resolution, visually complex digital environments. This paper introduces Iris, a foundational visual agent that addresses these challenges through two key innovations: Information-Sensitive Cropping (ISC) and Self-Refining Dual Learning (SRDL). ISC dynamically identifies and prioritizes visually dense regions using a edge detection algorithm, enabling efficient processing by allocating more computational resources to areas with higher information density. SRDL enhances the agent's ability to handle complex tasks by leveraging a dual-learning loop, where improvements in referring (describing UI elements) reinforce grounding (locating elements) and vice versa, all without requiring additional annotated data. Empirical evaluations demonstrate that Iris achieves state-of-the-art performance across multiple benchmarks with only 850K GUI annotations, outperforming methods using 10x more training data. These improvements further translate to significant gains in both web and OS agent downstream tasks.

Instruction tuning fine-tunes pre-trained Multi-modal Large Language Models (MLLMs) to handle real-world tasks. However, the rapid expansion of visual instruction datasets introduces data redundancy, leading to excessive computational costs. We propose a collaborative framework, DataTailor, which leverages three key principles--informativeness, uniqueness, and representativeness--for effective data selection. We argue that a valuable sample should be informative of the task, non-redundant, and represent the sample distribution (i.e., not an outlier). We further propose practical ways to score against each principle, which automatically adapts to a given dataset without tedious hyperparameter tuning. Comprehensive experiments on various benchmarks demonstrate that DataTailor achieves 101.3% of the performance of full-data fine-tuning with only 15% of the data, significantly reducing computational costs while maintaining superior results. This exemplifies the "Less is More" philosophy in MLLM development. The code and data is available in this \href{https://github.com/Yuqifan1117/DataTailor}{URL}.

The Multimodal Large Language Model (MLLM) is currently experiencing rapid growth, driven by the advanced capabilities of language-based LLMs. Unlike their specialist predecessors, existing MLLMs are evolving towards a Multimodal Generalist paradigm. Initially limited to understanding multiple modalities, these models have advanced to not only comprehend but also generate across modalities. Their capabilities have expanded from coarse-grained to fine-grained multimodal understanding and from supporting singular modalities to accommodating a wide array of or even arbitrary modalities. To assess the capabilities of various MLLMs, a diverse array of benchmark test sets has been proposed. This leads to a critical question: Can we simply assume that higher performance across tasks indicates a stronger MLLM capability, bringing us closer to human-level AI? We argue that the answer is not as straightforward as it seems. In this project, we introduce an evaluation framework to delineate the capabilities and behaviors of current multimodal generalists. This framework, named General-Level, establishes 5-scale levels of MLLM performance and generality, offering a methodology to compare MLLMs and gauge the progress of existing systems towards more robust multimodal generalists and, ultimately, towards AGI (Artificial General Intelligence). Central to our framework is the use of Synergy as the evaluative criterion, categorizing capabilities based on whether MLLMs preserve synergy across comprehension and generation, as well as across multimodal interactions. To evaluate the comprehensive abilities of various generalists, we present a massive multimodal benchmark, General-Bench, which encompasses a broader spectrum of skills, modalities, formats, and capabilities, including over 700 tasks and 325,800 instances. The evaluation results that involve over 100 existing state-of-the-art MLLMs uncover the capability rankings of generalists, highlighting the challenges in reaching genuine AI. We expect this project to pave the way for future research on next-generation multimodal foundation models, providing a robust infrastructure to accelerate the realization of AGI. Project Page: https://generalist.top/, Leaderboard: https://generalist.top/leaderboard/, Benchmark: https://huggingface.co/General-Level/.

As multimodal large language models (MLLMs) advance, MLLM-based virtual agents have demonstrated remarkable performance. However, existing benchmarks face significant limitations, including uncontrollable task complexity, extensive manual annotation, and a lack of multidimensional evaluation. In response to these challenges, we introduce OmniBench, a self-generating, graph-based benchmark with an automated pipeline for synthesizing tasks of controllable complexity through subtask composition. To evaluate the diverse capabilities of virtual agents on the graph, we further present OmniEval, a multidimensional evaluation framework that includes subtask-level evaluation, graph-based metrics, and comprehensive tests across 10 capabilities. Our synthesized dataset contains 36k graph-structured tasks across 20 scenarios, achieving a 91% human acceptance rate. Training on our graph-structured data shows that it improves generalization across environments. We conduct multidimensional evaluations for virtual agents, revealing their performance across various capabilities and paving the way for future advancements. Our project is available at https://omni-bench.github.io.

We present **HealthGPT**, a powerful Medical Large Vision-Language Model (Med-LVLM) that integrates medical visual comprehension and generation capabilities within a unified autoregressive paradigm. Our bootstrapping philosophy is to progressively adapt heterogeneous comprehension and generation knowledge to pre-trained Large Language Models (LLMs). This is achieved through a novel heterogeneous low-rank adaptation **(H-LoRA)** technique, which is complemented by a tailored hierarchical visual perception **(HVP)** approach and a three-stage learning strategy **(TLS)**. To effectively learn the HealthGPT, we devise a comprehensive medical domain-specific comprehension and generation dataset called **VL-Health**. Experimental results demonstrate exceptional performance and scalability of HealthGPT in medical visual unified tasks. Our project can be accessed at https://github.com/DCDmllm/HealthGPT.

The development of Generalist Virtual Agents (GVAs) has shown significant promise in autonomous task execution. However, current training paradigms face critical limitations, including reliance on outcome supervision and labor-intensive human annotations. To address these challenges, we propose Similar, a step-wise multi-dimensional generalist reward model, which offers fine-grained signals for agent training and can choose better actions for inference-time scaling. Specifically, we begin by systematically defining five dimensions for evaluating agent actions. Building on this framework, we design an MCTS-P algorithm to automatically collect and annotate step-wise, five-dimensional agent execution data. Using this data, we train Similar with our crafted Triple-M strategy. Furthermore, we introduce the first benchmark in the virtual agent domain for step-wise, multi-dimensional reward model training and evaluation, named SRM. This benchmark consists of two components: SRMTrain, which serves as the training set for Similar, and SRMEval, a manually selected test set for evaluating the reward model. Experimental results demonstrate that Similar, through its step-wise, multi-dimensional assessment and synergistic gain, provides GVAs with effective intermediate signals during both training and inference-time scaling. The code is available at https://github.com/antgroup/Similar.

Large language models (LLMs) have garnered increasing attention owing to their powerful comprehension and generation capabilities. Generally, larger LLMs (L-LLMs) that require paid interfaces exhibit significantly superior performance compared to smaller LLMs (S-LLMs) that can be deployed on a variety of devices. Knowledge distillation (KD) aims to empower S-LLMs with the capabilities of L-LLMs, while S-LLMs merely mimic the outputs of L-LLMs, failing to get the powerful decision-making capability for new situations. Consequently, S-LLMs are helpless when it comes to continuous decision-making tasks that require logical reasoning. To tackle the identified challenges, we propose a novel framework called Logic Distillation (LD). Initially, LD employs L-LLMs to instantiate complex instructions into discrete functions and illustrates their usage to establish a function base. Subsequently, LD fine-tunes S-LLMs based on the function base to learn the logic employed by L-LLMs in decision-making. During testing, S-LLMs will yield decision-making outcomes, function by function, based on current states. Experiments demonstrate that with the assistance of LD, S-LLMs can achieve outstanding results in continuous decision-making tasks, comparable to, or even surpassing, those of L-LLMs. The code and data for the proposed method are provided for research purposes https://github.com/Anfeather/Logic-Distillation.

In the fields of medicine and science, the volume of specialized literature has grown exponentially, containing vast multimodal data-text, images, and tables-that is essential for conveying in-depth scientific insights. However, effectively retrieving, processing, and answering high-level, complex queries from this data remains a significant challenge. In this study, we introduce MedQuery, a multimodal medical knowledge query-answering system that integrates query-based literature retrieval with a response generation module capable of reasoning across multimodal data. Our system begins with literature retrieval from PubMed, using keyword extraction and query alignment to improve document accuracy and relevance. Next, the multimodal processing module processes source documents, extracting images, tables, and text, converting all into a unified textual format. This data is then structured into a global graph capturing relationships among document elements, allowing our system to support a more integrated, in-depth understanding of complex medical queries beyond basic fact retrieval. Extensive evaluations across multiple datasets, including PubMedQA, PubMed-Summarization, and our own MedInquiry dataset, demonstrate that MedQuery outperforms traditional methods and existing commercial AI systems, achieving around 90% win rates in answer quality and a 13-36% improvement in accuracy.

In recent years, multimodal large models in the medical field have garnered widespread attention. However, this focus has primarily been on CT and MRI imaging, inadvertently neglecting the needs of economically underdeveloped regions and specific populations, such as pregnant women. These groups are often unable to utilize CT and MRI due to their prohibitive costs and potential harm to the body. Meanwhile, ultrasound, an economically viable and very low side effects medical imaging technique, has been largely overlooked by researchers. This study introduces a high-quality instruction-tuned Large vision Language Assistant for UltraSound (LLAUS), designed to answer questions about medical ultrasound images, aiming to assist clinicians in impoverished areas to improve the provision of healthcare services. To address the challenge of missing high-quality ultrasound data, we propose the Adaptive Caption Enhancement(ACE) and Adaptive Caption Optimization (ACO) strategies and have developed a high-quality instruction-following dataset. Subsequently, we fine-tune a Large Vision-Language Model (LVLM) using a novel Zoom-In method. By training on high-quality instruction-following datas, LLAUS demonstrates exceptional multimodal ultrasound communication capabilities, assisting in querying ultrasound images based on open-ended instructions. On tasks related to question-answering and caption generation for ultrasound images, LLAUS exhibits strong performance.

Multimodal Large Language Models (MLLMs) have garnered significant attention for their strong visual-semantic understanding. Most existing chart benchmarks evaluate MLLMs' ability to parse information from charts to answer questions. However, they overlook the inherent output biases of MLLMs, where models rely on their parametric memory to answer questions rather than genuinely understanding the chart content. To address this limitation, we introduce a novel Chart Hypothetical Question Answering (HQA) task, which imposes assumptions on the same question to compel models to engage in counterfactual reasoning based on the chart content. Furthermore, we introduce HAI, a human-AI interactive data synthesis approach that leverages the efficient text-editing capabilities of LLMs alongside human expert knowledge to generate diverse and high-quality HQA data at a low cost. Using HAI, we construct Chart-HQA, a challenging benchmark synthesized from publicly available data sources. Evaluation results on 18 MLLMs of varying model sizes reveal that current models face significant generalization challenges and exhibit imbalanced reasoning performance on the HQA task. Our codebase and newly generated datasets are available at https://github.com/chenxn2020/Chart-HQA.

Multimodal Industrial Anomaly Detection (MIAD)-fusing 3D point clouds and 2D RGB for product defect detection-is critical to quality inspection. However, existing MIAD methods assume all modalities are available and paired, overlooking real-scenario modality-missing and risking overfitting to incomplete data. To address these, we conduct the first comprehensive study on Modality-Incomplete Industrial Anomaly Detection (MIIAD) and establish MIIAD Bench , a benchmark covering diverse missing settings. Meanwhile, we propose RADAR, a robust two-stage Robust modAlity-instructive fusing & Detecting frAmewoRk. RADAR integrates i) a Modality-Incomplete Instruction mechanism-guiding the multimodal Transformer to focus more on available modal info, and ii) a Double-Pseudo Hybrid Module to highlight unique modality combinations and reduce overfitting. Our results show RADAR outperforms prior methods markedly on MIIAD Bench.

Semantic retrieval is crucial for modern applications yet remains underexplored in current research. Existing datasets are limited to single languages, single images, or singular retrieval conditions, often failing to fully exploit the expressive capacity of visual information as evidenced by maintained performance when images are replaced with captions. However, practical retrieval scenarios frequently involve interleaved multi-condition queries with multiple images. Hence, this paper introduces MERIT, the first multilingual dataset for interleaved multi-condition semantic retrieval, comprising 320,000 queries with 135,000 products in 5 languages, covering 7 distinct product categories. Extensive experiments on MERIT identify existing models's critical limitation: focusing solely on global semantic information while neglecting specific conditional elements in queries. Consequently, we propose Coral, a novel fine-tuning framework that adapts pre-trained MLLMs by integrating embedding reconstruction to preserve fine-grained conditional elements and contrastive learning to extract comprehensive global semantics. Experiments demonstrate that Coral achieves a 45.9% performance improvement over conventional approaches on MERIT, with strong generalization capabilities validated across 8 established retrieval benchmarks. Collectively, our contributions—a novel dataset, identification of critical limitations in existing approaches, and an innovative fine-tuning framework—establish a foundation for future research in interleaved multi-condition semantic retrieval. **Data & Code**: [MERIT-2025.github.io](https://merit-2025.github.io)

The rapid development of Multimodal Large Language Models (MLLMs) poses increasing demands on the diversity and complexity of multimodal datasets. Yet manual annotation pipelines can no longer keep pace. Existing augmentation methods often follow fixed rules and lack verifiable control over sample diversity and reasoning complexity. To address this, we introduce Scalable COunterfactual Program Evolution (SCOPE), a framework that uses symbolic Visual Programming to guide program evolution via counterfactual reasoning. SCOPE performs the three steps of counterfactual inference: (1) Abduction, by generating verifiable programs to model reasoning associations; (2) Action, by intervening on program structure along three axes—reasoning path, visual context, and cross-instance composition; and (3) Prediction, by categorizing evolved instances by difficulty, structure, and input multiplicity. Based on this process, we build SCOPE-Train and SCOPE-Test, evolving benchmarks with expert validation. To support training, we propose MAP, a curriculum learning strategy that aligns model capacity with sample difficulty. Experiments show that SCOPE improves reasoning performance, exposes model blind spots, and enhances visual dialog capabilities.

Recent endeavors in Multimodal Large Language Models (MLLMs) aim to unify visual comprehension and generation. However, these two capabilities remain largely independent, as if they are two separate functions encapsulated within the same model. Consequently, visual comprehension does not enhance visual generation, and the reasoning mechanisms of LLMs have not been fully integrated to revolutionize image generation. In this paper, we propose to enable the collaborative co-evolution of visual comprehension and generation, advancing image generation into an iterative introspective process. We introduce a two-stage training approach: supervised fine-tuning teaches the MLLM with the foundational ability to generate genuine CoT for visual generation, while reinforcement learning activates its full potential via an exploration-exploitation trade-off. Ultimately, we unlock the Aha moment in visual generation, advancing MLLMs from text-to-image tasks to unified image generation. Extensive experiments demonstrate that our model not only excels in text-to-image generation and image editing, but also functions as a superior image semantic evaluator with enhanced visual comprehension capabilities. Project Page: \url{https://janus-pro-r1.github.io}.

Recent advances in reinforcement learning (RL) methods such as Grouped Relative Policy Optimization (GRPO) have strengthened the reasoning capabilities of Large Vision-Language Models (LVLMs). However, due to the inherent entanglement between visual and textual modalities, applying GRPO to LVLMs often leads to reward convergence across different responses to the same sample as training progresses, hindering effective gradient updates and causing the enhancement of chain-of-thought reasoning to stagnate or even collapse. To address this issue, we propose a progressive instruction evolution framework, EvolvedGRPO, to gradually generate more complex questions via editing instructions in an adversarial way, progressively aligned with the model’s evolving capabilities. Specifically, we design two instruction editing strategies across modalities, incorporating incrementally increasing editing instructions and RL-based adversarial data augmentation to improve the effectiveness of model training. To address GRPO's limitations on overly difficult problems, we first train on basic subproblem versions of complex multi-modal questions in both the visual and textual modalities, progressively increasing difficulty to enable prefix-style process rewards, effectively combining the strengths of both process rewards and group-wise relative rewards. Finally, EvolvedGRPO achieves state-of-the-art performance among open-source RL models on multi-modal reasoning tasks, even approaching the closed-source GPT-4o in reasoning capabilities, and demonstrates better performance on unseen LVLM general benchmarks. The Code for EvolvedGRPO is available at https://github.com/SHENZHEBEI/EvolvedGRPO.

Large language models (LLMs) have shown promise in empowering robotics, but their widespread real-world application remains challenging due to two main issues: (1) existing research is "out-of-the-box", struggling to generalize to new robots and tasks, especially long-horizon tasks, and (2) deploying more general and powerful LLMs exceeds the capabilities of commodity hardware. To address these challenges, we propose the edge-cloud collaborative framework, i.e., SkyNet. We deploy LLMs in the cloud to create initial plans, select executable skills from a predefined library, and send them to the edge-based robot. The robot integrates multiple modules to form a policy network to complete the skills and update the feedback history. Based on the feedback history, the cloud LLMs determine whether to replan. The edge-cloud collaborative approach alleviates the pressure of deploying LLMs on commodity hardware, while the modular design enables easy extension to different tasks or robots without reconfiguring everything. To address the lack of standardized real-world experimental setups, we set up two easily replicable long-horizon tasks on a mobile robot equipped with commodity hardware, analyze the performance of various modules, and demonstrated the effectiveness of SkyNet.

Recently, research on Text-Attributed Graphs (TAGs) has gained significant attention due to the prevalence of free-text node features in real-world applications and the advancements in Large Language Models (LLMs) that bolster TAG methodologies. However, current TAG approaches face two primary challenges: (i) Heavy reliance on label information and (ii) Limited cross-domain zero/few-shot transferability. These issues constrain the scaling of both data and model size, owing to high labor costs and scaling laws, complicating the development of graph foundation models with strong transferability. In this work, we propose the GraphCLIP framework to address these challenges by learning graph foundation models with strong cross-domain zero/few-shot transferability through a self-supervised contrastive graph-summary pretraining method. Specifically, we generate and curate large-scale graph-summary pair data with the assistance of LLMs, and introduce a novel graph-summary pretraining method, combined with invariant learning, to enhance graph foundation models with strong cross-domain zero-shot transferability. For few-shot learning, we propose a novel graph prompt tuning technique aligned with our pretraining objective to mitigate catastrophic forgetting and minimize learning costs. Extensive experiments show the superiority of GraphCLIP in both zero-shot and few-shot settings, while evaluations across various downstream tasks confirm the versatility of GraphCLIP. Our code is available at: https://github.com/ZhuYun97/GraphCLIP.

Continual learning (CL) has emerged as a crucial paradigm for learning from sequential data while retaining previous knowledge. Continual graph learning (CGL), characterized by dynamically evolving graphs from streaming data, presents distinct challenges that demand efficient algorithms to prevent catastrophic forgetting. The first challenge stems from the interdependencies between different graph data, in which previous graphs influence new data distributions. The second challenge is handling large graphs in an efficient manner. To address these challenges, we propose an efficient continual graph learner (E-CGL) in this paper. We address the interdependence issue by demonstrating the effectiveness of replay strategies and introducing a combined sampling approach that considers both node importance and diversity. To improve efficiency, E-CGL leverages a simple yet effective multilayer perceptron (MLP) model that shares weights with a graph neural network (GNN) during training, thereby accelerating computation by circumventing the expensive message-passing process. Our method achieves state-of-the-art results on four CGL datasets under two settings, while significantly lowering the catastrophic forgetting value to an average of −1.1%. Additionally, E-CGL achieves the training and inference speedup by an average of 15.83× and 4.89×, respectively, across four datasets. These results indicate that E-CGL not only effectively manages correlations between different graph data during continual training but also enhances efficiency in large-scale CGL.

Recent years have seen a surge of interest in anomaly detection. However, existing unsupervised anomaly detectors, particularly those for the vision modality, face significant challenges due to redundant information and sparse latent space. In contrast, anomaly detectors demonstrate superior performance in the language modality due to the unimodal nature of the data. This paper tackles the aforementioned challenges for vision modality from a multimodal point of view. Specifically, we propose Cross-modal Guidance (CMG), comprising of Cross-modal Entropy Reduction (CMER) and Cross-modal Linear Embedding (CMLE), to address the issues of redundant information and sparse latent space, respectively. CMER involves masking portions of the raw image and computing the matching score with the corresponding text. Essentially, CMER eliminates irrelevant pixels to direct the detector's focus towards critical content. To learn a more compact latent space for the vision anomaly detection, CMLE learns a correlation structure matrix from the language modality. Then, the acquired matrix compels the distribution of images to resemble that of texts in the latent space. Extensive experiments demonstrate the effectiveness of the proposed methods. Particularly, compared to the baseline that only utilizes images, the performance of CMG has been improved by 16.81%. Ablation experiments further confirm the synergy among the proposed CMER and CMLE, as each component depends on the other to achieve optimal performance.

Diffusion models have demonstrated remarkable capabilities for text-to-video (T2V) editing tasks, relying on fine-tuning for pretrained text-to-image (T2I) diffusion models with only one video-prompt pair. However, conventional fine-tuning approaches require tuning and storing numerous parameters for each video, leading to substantial parameter and memory costs. To mitigate these issues, we propose Truncating Diffusion, an efficient fine-tuning method for video editing that optimizes both parameter and memory usage. Specifically, we propose the Truncating Diffusion module, which is designed with a focus on module architecture and initialization, specifically targeting optimization with a small training set, such as a single video-prompt pair. Theoretical analysis using the Johnson-Lindenstrauss lemma and the Eckart-Young-Mirsky theorem shows that Truncating Diffusion can achieve a minimal Frobenius norm distance to the original attention algorithm with appropriate initialization, which enhances the ease of optimization and improves video editing performance. During fine-tuning, the Truncating Diffusion module integrates seamlessly with the original diffusion model. We freeze the weights of the denoising network within the original pretrained diffusion model, updating only the introduced low-rank alternatives to ensure parameter and memory efficiency. Additionally, we propose Latent Flow Loss and Bidirectional Inter-Frame Attention (BIFA) to improve temporal consistency in synthesized videos. The Latent Flow Loss leverages global temporal information from the input video during training, while BIFA utilizes local temporal information from adjacent frames during inference. These enhancements do not incur additional memory or parameter costs during fine-tuning. Comparisons with state-of-the-art approaches demonstrate that Truncating Diffusion provides superior text alignment, video quality, and inter-frame temporal consistency in video editing. Importantly, Truncating Diffusion requires fine-tuning only 3.2% of the parameters and uses just 62% of the memory compared to the baseline model.

With the increasing demand for data privacy, federated learning (FL) has gained popularity for various applications. Most existing FL works focus on the classification task, overlooking those scenarios where anomaly detection may also require privacy-preserving. Traditional anomaly detection algorithms cannot be directly applied to the FL setting due to false and missing detection issues. Moreover, with common aggregation methods used in FL (e.g., averaging model parameters), the global model cannot keep the capacities of local models in discriminating anomalies deviating from local distributions, which further degrades the performance. For the aforementioned challenges, we propose Federated Anomaly Detection with Noisy Global Density Estimation, and Self-supervised Ensemble Distillation (FADngs). Specifically, FADngs aligns the knowledge of data distributions from each client by sharing processed density functions. Besides, FADngs trains local models in an improved contrastive learning way that learns more discriminative representations specific for anomaly detection based on the shared density functions. Furthermore, FADngs aggregates capacities by ensemble distillation, which distills the knowledge learned from different distributions to the global model. Our experiments demonstrate that the proposed method significantly outperforms state-of-the-art federated anomaly detection methods. We also empirically show that the shared density function is privacy-preserving. The code for the proposed method is provided for research purposes https://github.com/kanade00/Federated_Anomaly_detection.

Many studies have aimed to improve Transformer model efficiency using low-rank-based methods that compress sequence length with predetermined or learned compression matrices. However, these methods fix compression coefficients for tokens in the same position during inference, ignoring sequence-specific variations. They also overlook the impact of hidden state dimensions on efficiency gains. To address these limitations, we propose dynamic bilinear low-rank attention (DBA), an efficient and effective attention mechanism that compresses sequence length using input-sensitive dynamic compression matrices. DBA achieves linear time and space complexity by jointly optimizing sequence length and hidden state dimension while maintaining state-of-the-art performance. Specifically, we demonstrate through experiments and the properties of low-rank matrices that sequence length can be compressed with compression coefficients dynamically determined by the input sequence. In addition, we illustrate that the hidden state dimension can be approximated by extending the Johnson-Lindenstrauss lemma, thereby introducing only a small amount of error. DBA optimizes the attention mechanism through bilinear forms that consider both the sequence length and hidden state dimension. Moreover, the theoretical analysis substantiates that DBA excels at capturing high-order relationships in cross-attention problems. Experimental results across different tasks with varied sequence length conditions demonstrate that DBA achieves state-of-the-art performance compared to several robust baselines. DBA also maintains higher processing speed and lower memory usage, highlighting its efficiency and effectiveness across diverse applications.

Pretrained large models, particularly large language models, have garnered increasing attention, as they have demonstrated remarkable abilities through contextual learning. Pretrained large models are increasingly recognized as fundamental tools for solving various tasks. However, the substantial computational demands of large models have dissuaded most product teams and individuals from running them. In such scenarios, to leverage the exceptional performance of large models, one must solely depend on costly APIs, further burdening product teams and individuals. On the other hand, despite the overall inferior performance of small models compared to large models, there are certain distributions where small models can achieve comparable or even superior results. For instance, during training, small models may become trapped in a local optimum that is unique to certain distributions, leading to superior performance. Hence, we propose Data Shunt (DS), a general paradigm for collaboration of small and large models. DS not only substantially reduces the cost associated with deploying large models but also effectively enhances overall performance. Specifically, DS determines the shunting direction by evaluating the confidence level of small models. When the confidence level falls below a specific threshold, the input data is forwarded to large models. To further leverage the advantages of the small and large models, we introduce Prompt Pruning (PP) and 2-Stage Confidence Distillation (2CD), which facilitate mutual collaboration, leading to better results and less cost. The remarkable performance across diverse modalities and tasks demonstrates the superiority of the proposed DS over large models. For instance, ChatGPT achieves an accuracy of 94.43% on Amazon Product sentiment analysis, and DS achieves an accuracy of 95.64%, while the cost has been reduced to only 31.18%. The code for the proposed method are provided for research purposes https://github.com/Anfeather/Data-Shunt.

In this work, we propose a two-stage sign language production (SLP) paradigm that first encodes sign language sequences into discrete codes and then autoregressively generates sign language from text based on the learned codebook. However, existing vector quantization (VQ) methods are fixed-length encodings, overlooking the uneven information density in sign language, which leads to under-encoding of important regions and over-encoding of unimportant regions. To address this issue, we propose a novel dynamic vector quantization (DVA-VAE) model that can dynamically adjust the encoding length based on the information density in sign language to achieve accurate and compact encoding. Then, a GPT-like model learns to generate code sequences and their corresponding durations from spoken language text. Extensive experiments conducted on the PHOENIX14T dataset demonstrate the effectiveness of our proposed method. To promote sign language research, we propose a new large German sign language dataset, PHOENIX-News, which contains 486 hours of sign language videos, audio, and transcription texts. Experimental analysis on PHOENIX-News shows that the performance of our model can be further improved by increasing the size of the training data. Our project homepage is https://t2sgpt-demo.yinaoxiong.cn.

In image question answering, due to the abundant and sometimes redundant information, precisely matching and integrating the information from both text and images is a challenge. In this paper, we propose the Decomposition-Integration Enhancing Multimodal Insight (DIEM) which initially decomposes the given question and image into multiple subquestions and several sub-images aiming to isolate specific elements for more focused analysis. We then in-tegrate these sub-elements by matching each subquestion with its relevant sub-images, while also retaining the original image, to construct a comprehensive answer to the original question without losing sight of the overall context. This strategy mirrors the human cognitive process of simplifying complex problems into smaller components for individual analysis, followed by an integration of these insights. We implement DIEM on the LLaVA-v1.5 model, and evaluate its performance on ScienceQA and MM-Vet. Ex-perimental results indicate that our method boosts accu-racy in most question classes of the ScienceQA (+2.03% in average), especially in the image modality (+3.40%). On MM-Vet, our method achieves an improvement in MM-Vet scores, increasing from 31.1 to 32.4. These findings high-light DIEM's effectiveness in harmonizing the complexities of multimodal data, demonstrating its ability to enhance accuracy and depth in image question answering through its decomposition-integration process.

Multi-modal Large Language Models (MLLMs) tuned on machine-generated instruction-following data have demonstrated remarkable performance in various multi-modal understanding and generation tasks. However, the hallucinations inherent in machine-generated data, which could lead to hallucinatory outputs in MLLMs, remain under-explored. This work aims to investigate various hallucinations (i.e., object, relation, attribute hallucinations) and mitigate those hallucinatory toxicities in large-scale machine-generated visual instruction datasets. Drawing on the human ability to identify factual errors, we present a novel hallucination detection and elimination framework, HalluciDoctor, based on the cross-checking paradigm. We use our framework to identify and eliminate hallucinations in the training data automatically. Interestingly, HalluciDoctor also indicates that spurious correlations arising from long-tail object cooccurrences contribute to hallucinations. Based on that, we execute counterfactual visual instruction expansion to balance data distribution, thereby enhancing MLLMs' resistance to hallucinations. Comprehensive experiments on hallucination evaluation benchmarks show that our method successfully mitigates 44.6% hallucinations relatively and maintains competitive performance compared to LLaVA. The data and code for this paper are publicly available.11https://github.com/Yuqifan1117/HalluciDoctor

Representation learning on text-attributed graphs (TAGs) is vital for real-world applications, as they combine semantic textual and contextual structural information.Research in this field generally consist of two main perspectives: local-level encoding and global-level aggregating, respectively refer to textual node information unification (e.g., using Language Models) and structure-augmented modeling (e.g., using Graph Neural Networks).Most existing works focus on combining different information levels but overlook the interconnections, i.e., the contextual textual information among nodes, which provides semantic insights to bridge local and global levels.In this paper, we propose GraphBridge, a multi-granularity integration framework that bridges local and global perspectives by leveraging contextual textual information, enhancing fine-grained understanding of TAGs.Besides, to tackle scalability and efficiency challenges, we introduce a graph-aware token reduction module.Extensive experiments across various models and datasets show that our method achieves state-of-the-art performance, while our graph-aware token reduction module significantly enhances efficiency and solves scalability issues.Codes are available at https://github.com/wykk00/GraphBridge

The primary objective of simultaneous machine translation (SiMT) is to minimize latency while preserving the quality of the final translation. Drawing inspiration from CPU branch prediction techniques, we propose incorporating branch prediction techniques in SiMT tasks to reduce translation latency. Specifically, we utilize a language model as a branch predictor to predict potential branch directions, namely, future source words. Subsequently, we utilize the predicted source words to decode the output in advance. When the actual source word deviates from the predicted source word, we use the real source word to decode the output again, replacing the predicted output. To further reduce computational costs, we share the parameters of the encoder and the branch predictor, and utilize a pre-trained language model for initialization. Our proposed method can be seamlessly integrated with any SiMT model. Extensive experimental results demonstrate that our approach can improve translation quality and latency at the same time. Our code is available at https://github.com/YinAoXiong/simt_branch_predictor.

Recent advancements in Multimodal Large Language Models (MLLMs) have been utilizing Visual Prompt Generators (VPGs) to convert visual features into tokens that LLMs can recognize. This is achieved by training the VPGs on millions of image-caption pairs, where the VPG-generated tokens of images are fed into a frozen LLM to generate the corresponding captions. However, this image-captioning based training objective inherently biases the VPG to concentrate solely on the primary visual contents sufficient for caption generation, often neglecting other visual details. This shortcoming results in MLLMs’ underperformance in comprehending demonstrative instructions consisting of multiple, interleaved, and multimodal instructions that demonstrate the required context to complete a task. To address this issue, we introduce a generic and lightweight Visual Prompt Generator Complete module (VPG-C), which can infer and complete the missing details essential for comprehending demonstrative instructions. Further, we propose a synthetic discriminative training strategy to fine-tune VPG-C, eliminating the need for supervised demonstrative instructions. As for evaluation, we build DEMON, a comprehensive benchmark for demonstrative instruction understanding. Synthetically trained with the proposed strategy, VPG-C achieves significantly stronger zero-shot performance across all tasks of DEMON. Further evaluation on the MME and OwlEval benchmarks also demonstrate the superiority of VPG-C. The code and models are available at https://github.com/DCDmllm/Cheetah.

We introduce InstructVid2Vid, an end-to-end diffusion-based methodology for video editing guided by human language instructions. Our approach empowers video manipulation guided by natural language directives, eliminating the need for per-example fine-tuning or inversion. The proposed InstructVid2Vid model modifies a pretrained image generation model, Stable Diffusion, to generate a time-dependent sequence of video frames. By harnessing the collective intelligence of disparate models, we engineer a training dataset rich in video-instruction triplets, which is a more cost-efficient alternative to collecting data in real-world scenarios. To enhance the coherence between successive frames within the generated videos, we propose the Inter-Frames Consistency Loss and incorporate it during the training process. With multimodal classifier-free guidance during the inference stage, the generated videos is able to resonate with both the input video and the accompanying instructions. Experimental results demonstrate that InstructVid2Vid is capable of generating high-quality, temporally coherent videos and performing diverse edits, including attribute editing, background changes, and style transfer. These results underscore the versatility and effectiveness of our proposed method.

While recent large-scale text-to-speech (TTS) models have achieved significant progress, they still fall shorts in speech quality, similarity, and prosody. Considering that speech intricately encompasses various attributes (e.g., content, prosody, timbre, and acoustic details) that pose significant challenges for generation, a natural idea is to factorize speech into individual subspaces representing different attributes and generate them individually. Motivated by it, we propose a TTS system with novel factorized diffusion models to generate natural speech in a zero-shot way. Specifically, 1) we design a neural codec with factorized vector quantization (FVQ) to disentangle speech waveform into subspaces of content, prosody, timbre, and acoustic details; 2) we propose a factorized diffusion model, which generates attributes in each subspace following its corresponding prompt. With this factorization design, our method can effectively and efficiently model the intricate speech with disentangled subspaces in a divide-and-conquer way. Experimental results show that our method outperforms the state-of-the-art TTS systems on quality, similarity, prosody, and intelligibility.

For multimodal LLMs, the synergy of visual comprehension (textual output) and generation (visual output) presents an ongoing challenge. This is due to a conflicting objective: for comprehension, an MLLM needs to abstract the visuals; for generation, it needs to preserve the visuals as much as possible. Thus, the objective is a dilemma for visual-tokens. To resolve the conflict, we propose encoding images into morph-tokens to serve a dual purpose: for comprehension, they act as visual prompts instructing MLLM to generate texts; for generation, they take on a different, non-conflicting role as complete visual-tokens for image reconstruction, where the missing visual cues are recovered by the MLLM. Extensive experiments show that morph-tokens can achieve a new SOTA for multimodal comprehension and generation simultaneously. Our project is available at https://github.com/DCDmllm/MorphTokens.

Large Language Models (LLMs) demonstrate remarkable proficiency in comprehending and handling text-based tasks. Many efforts are being made to transfer these attributes to video modality, which are termed Video-LLMs. However, existing Video-LLMs can only capture the coarse-grained semantics and are unable to effectively handle tasks related to comprehension or localization of specific video segments. In light of these challenges, we propose Momentor, a Video-LLM capable of accomplishing fine-grained temporal understanding tasks. To support the training of Momentor, we design an automatic data generation engine to construct Moment-10M, a large-scale video instruction dataset with segment-level instruction data. We train Momentor on Moment-10M, enabling it to perform segment-level reasoning and localization. Zero-shot evaluations on several tasks demonstrate that Momentor excels in fine-grained temporally grounded comprehension and localization.

Visual programming, a modular paradigm, integrates different modules and Python operators to solve various vision-language tasks. Unlike end-to-end models that need task-specific data, it performs visual processing and inference in an unsupervised manner. Current visual programming methods generate programs in a single pass where the ability to evaluate and optimize based on feedback, unfortunately, is lacking, which consequentially limits their effectiveness for complex, multi-step problems. Inspired by benders decomposition, we introduce De-fine, a training-free framework that automatically decomposes complex tasks into simpler subtasks and refines programs through auto-feedback. This model-agnostic approach can improve logical reasoning performance by integrating the strengths of multiple models. Our experiments across various visual tasks show that De-fine creates more robust programs. Moreover, viewing each feedback module as an independent agent will yield fresh prospects for the field of agent research.

The remarkable performance of Multimodal Large Language Models (MLLMs) has demonstrated their proficient understanding capabilities in handling various visual tasks. Nevertheless, the opaque nature of black-box reasoning processes persists as an enigma, rendering them uninterpretable and struggling with hallucination. Their ability to execute intricate reasoning tasks is also constrained, culminating in stagnation of progression. In this work, we introduce Fact, a novel paradigm designed to generate multimodal rationales that are faithful, concise, and transferable for teaching MLLMs. This paradigm utilizes verifiable visual programming to generate executable code guaranteeing faithfulness. Through a series of operations including pruning, merging, and bridging, the rationale enhances its conciseness. Furthermore, we filter rationales that can be transferred to end-to-end paradigms from programming paradigms to guarantee transferability. Empirical evidence from experiments demonstrates the superiority of Fact across models of varying parameter sizes, significantly enhancing their compositional reasoning and generalization ability and reducing hallucinations owing to its high correlation between images and text.

We introduce the DEMON Challenge, defined as a benchmark for demonstrative instruction following, to ACM Multimedia 2024. The DEMON Challenge aims to assess the ability of models and systems to comprehend demonstrative instructions consisting of multiple, interleaved, and multimodal context that demonstrate the required information to complete a task. These instructions are curated from a diverse range of multi-modal datasets, spanning various fields and scenarios, to ensure comprehensive coverage and challenge diversity. The challenge details and participation information are available on the https://dcdmllm.github.io/DEMON-challenge/.

World models are progressively being employed across diverse fields, extending from basic environment simulation to complex scenario construction. However, existing models are mainly trained on domain-specific states and actions, and confined to single-modality state representations. In this paper, We introduce WorldGPT, a generalist world model built upon Multimodal Large Language Model (MLLM). WorldGPT acquires an understanding of world dynamics through analyzing millions of videos across various domains. To further enhance WorldGPT's capability in specialized scenarios and long-term tasks, we have integrated it with a novel cognitive architecture that combines memory offloading, knowledge retrieval, and context reflection. As for evaluation, we build WorldNet, a multimodal state transition prediction benchmark encompassing varied real-life scenarios. Conducting evaluations on WorldNet directly demonstrates WorldGPT's capability to accurately model state transition patterns, affirming its effectiveness in understanding and predicting the dynamics of complex scenarios. We further explore WorldGPT's emerging potential in serving as a world simulator, helping multimodal agents generalize to unfamiliar domains through efficiently synthesising multimodal instruction instances which are proved to be as reliable as authentic data for fine-tuning purposes. The code and dataset are available on the https://github.com/DCDmllm/WorldGPT

In recent times, Vision-Language Models (VLMs) have been trained under two predominant paradigms. Generative training has enabled Multimodal Large Language Models (MLLMs) to tackle various complex tasks, yet issues such as hallucinations and weak object discrimination persist. Discriminative training, exemplified by models like CLIP, excels in zero-shot image-text classification and retrieval, yet struggles with complex scenarios requiring fine-grained semantic differentiation. This paper addresses these challenges by proposing a unified approach that integrates the strengths of both paradigms. Considering interleaved image-text sequences as the general format of input samples, we introduce a structure-induced training strategy that imposes semantic relationships between input samples and the MLLM’s hidden state. This approach enhances the MLLM’s ability to capture global semantics and distinguish fine-grained semantics. By leveraging dynamic sequence alignment within the Dynamic Time Warping framework and integrating a novel kernel for fine-grained semantic differentiation, our method effectively balances generative and discriminative tasks. Extensive experiments demonstrate the effectiveness of our approach, achieving state-of-the-art results in multiple generative tasks, especially those requiring cognitive and discrimination abilities. Additionally, our method surpasses discriminative benchmarks in interleaved and fine-grained retrieval tasks. By employing a retrieval-augmented generation strategy, our approach further enhances performance in some generative tasks within one model, offering a promising direction for future research in vision-language modeling.

The swift advancement in Multimodal LLMs (MLLMs) also presents significant challenges for effective knowledge editing. Current methods, including intrinsic knowledge editing and external knowledge resorting, each possess strengths and weaknesses, struggling to balance the desired properties of reliability, generality, and locality when applied to MLLMs. In this paper, we propose \textbf{UniKE}, a novel multimodal editing method that establishes a unified perspective and paradigm for intrinsic knowledge editing and external knowledge resorting. Both types of knowledge are conceptualized as vectorized key-value memories, with the corresponding editing processes resembling the assimilation and accommodation phases of human cognition, conducted at the same semantic levels. Within such a unified framework, we further promote knowledge collaboration by disentangling the knowledge representations into the semantic and truthfulness spaces. Extensive experiments validate the effectiveness of our method, which ensures that the post-edit MLLM simultaneously maintains excellent reliability, generality, and locality. The code for UniKE is available at https://github.com/beepkh/UniKE.

Recent studies indicate that dense retrieval models struggle to perform well on a wide variety of retrieval tasks that lack dedicated training data, as different retrieval tasks often entail distinct search intents. To address this challenge, in this work we leverage instructions to flexibly describe retrieval intents and introduce I3, a unified retrieval system that performs Intent-Introspective retrieval across various tasks, conditioned on Instructions without any task-specific training. I3 innovatively incorporates a pluggable introspector in a parameter-isolated manner to comprehend specific retrieval intents by jointly reasoning over the input query and instruction, and seamlessly integrates the introspected intent into the original retrieval model for intent-aware retrieval. Furthermore, we propose progressively-pruned intent learning. It utilizes extensive LLM-generated data to train I3 phase-by-phase, embodying two key designs: progressive structure pruning and drawback extrapolation-based data refinement. Extensive experiments show that in the BEIR benchmark, I3 significantly outperforms baseline methods designed with task-specific retrievers, achieving state-of-the-art zero-shot performance without any task-specific tuning.

In this work, we investigate the problem of out-of-distribution (OOD) generalization for unsupervised learning methods on graph data. To improve the robustness against such distributional shifts, we propose a Model-Agnostic Recipe for Improving OOD generalizability of unsupervised graph contrastive learning methods, which we refer to as MARIO. MARIO introduces two principles aimed at developing distributional-shift-robust graph contrastive methods to overcome the limitations of existing frameworks: (i) Invariance principle that incorporates adversarial graph augmentation to obtain invariant representations and (ii) Information Bottleneck (IB) principle for achieving generalizable representations through refining representation contrasting. To the best of our knowledge, this is the first work that investigates the OOD generalization problem of graph contrastive learning, with a specific focus on node-level tasks. Through extensive experiments, we demonstrate that our method achieves state-of-the-art performance on the OOD test set, while maintaining comparable performance on the in-distribution test set when compared to existing approaches. Our codes are available at: https://github.com/ZhuYun97/MARIO.

Graph self-supervised algorithms have achieved significant success in acquiring generic knowledge from abundant unlabeled graph data. These pre-trained models can be applied to various downstream Web applications, saving training time and improving downstream performance. However, variations in attribute semantics across graphs pose challenges in transferring pre-trained models to downstream tasks. Concretely speaking, for example, the additional task-specific node information in downstream tasks (specificity) is usually deliberately omitted so that the pre-trained representation (transferability) can be leveraged. The trade-off as such is termed as "transferability-specificity dilemma" in this work. To address this challenge, we introduce an innovative deployment module coined as GraphControl, motivated by ControlNet, to realize better graph domain transfer learning. Specifically, by leveraging universal structural pre-trained models and GraphControl, we align the input space across various graphs and incorporate unique characteristics of target data as conditional inputs. These conditions will be progressively integrated into the model during fine-tuning or prompt tuning through ControlNet, facilitating personalized deployment. Extensive experiments show that our method significantly enhances the adaptability of pre-trained models on target attributed datasets, achieving 1.4-3x performance gain. Furthermore, it outperforms training-from-scratch methods on target data with a comparable margin and exhibits faster convergence. Our codes are available at: https://github.com/wykk00/GraphControl.

The visual question generation (VQG) task aims to generate human-like questions from an image and potentially other side information (e.g., answer type). Previous works on VQG fall in two aspects: i) They suffer from one image to many questions mapping problem, which leads to the failure of generating referential and meaningful questions from an image. ii) They fail to model complex implicit relations among the visual objects in an image and also overlook potential interactions between the side information and image. To address these limitations, we first propose a novel learning paradigm to generate visual questions with answer-awareness and region-reference. Concretely, we aim to ask the right visual questions with Double Hints - textual answers and visual regions of interests, which could effectively mitigate the existing one-to-many mapping issue. Particularly, we develop a simple methodology to self-learn the visual hints without introducing any additional human annotations. Furthermore, to capture these sophisticated relationships, we propose a new double-hints guided Graph-to-Sequence learning framework, which first models them as a dynamic graph and learns the implicit topology end-to-end, and then utilizes a graph-to-sequence model to generate the questions with double hints. Experimental results demonstrate the priority of our proposed method.

In the training procedures of many real-world learning models, gathering and annotating decent amounts of labeled data can be cost-prohibitive. To mitigate this data-hungry problem, active learning (AL) and semi-supervised learning (SSL) are frequently adopted as two effective but often isolated means. Some recent studies explored the potential of combining AL and SSL to better probe the unlabeled data. However, almost all these contemporary SSL-AL works use a simple combination strategy, ignoring SSL and AL's inherent relation. Further, other methods suffer from high computational costs when dealing with large-scale, high-dimensional datasets. Motivated by the industry practice of labeling data, we first propose an innovativeInconsistency-based virtual aDvErsarialActiveLearning (IDEAL) algorithm to further investigate SSL-AL's potential superiority and achieve mutual enhancement of AL and SSL, i.e., SSL propagates label information to unlabeled samples and provides smoothed embeddings for AL, while AL excludes samples with inconsistent predictions and considerable uncertainty for SSL. We estimate unlabeled samples’ inconsistency by augmentation strategies of different granularities, including fine-grained continuous perturbation exploration and coarse-grained data transformations. Moreover, to solve the problem that the utilization efficiency of unlabeled samples is still insufficient in the process of semi-supervised training, we extend our IDEAL to a curriculum-guided version, namely SPL-IDEAL algorithm. The SPL-IDEAL algorithm can regularize the training process towards better regions in parameter space and denoise the pseudo labels with low confidence, achieving better performance. The extensive experiments, in both text and image benchmark datasets, validate the effectiveness of our proposed IDEAL and SPL-IDEAL algorithms, comparing them against state-of-the-art baselines. Two real-world case studies visualize the practical industrial value of applying and deploying the proposed data sampling algorithms.

Dynamic graph-based data are ubiquitous in the real world, such as social networks, finance systems, and traffic flow. Fast and accurately detecting anomalies in these dynamic graphs is of vital importance. However, despite promising results the current anomaly detection methods have achieved, there are two major limitations when coping with dynamic graphs. The first limitation is that the topological structures and the temporal dynamics have been modeled separately, resulting in less expressive features for detection. The second limitation is that the models have been trained by unreliable noisy labels generated by random negative sampling, rendering it severely vulnerable to subtle perturbations. To overcome the above limitations, we propose RustGraph, a robust anomaly detection framework by jointly learning structural-temporal dependency in dynamic graphs. To this end, we design a variational graph auto-encoder with informative prior that simultaneously encodes both graph structural and temporal information. Then we introduce a fine-grained contrastive learning method to learn better node representations by utilizing the temporal consistency between two snapshots. Furthermore, we formulate the noisy label learning problem for anomaly detection in dynamic graph, and then propose a robust anomaly detector to improve the model performance by leveraging learned graph structure signal. Our extensive experiments on six real-world datasets demonstrate the proposed RustGraph method achieves state-of-the-art performance with an average of 3.64% improvement on AUC-ROC metric compared with all baselines. The codes are available athttps://github.com/aubreygjh/RustGraph.

Machine Learning is often challenged by insufficient labeled data. Previous methods employing implicit commonsense knowledge of pre-trained language models (PLMs) or pattern-based symbolic knowledge have achieved great success in mitigating manual annotation efforts. In this paper, we focus on the collaboration among different knowledge sources and present KICE, a Knowledge-evolving framework by Iterative Consolidation and Expansion with the guidance of PLMs and rule-based patterns. Specifically, starting with limited labeled data as seeds, KICE first builds a Rule Generator by prompt-tuning to stimulate the rich knowledge distributed in PLMs, generate seed rules, and initialize the rules set. Afterwards, based on the rule-labeled data, the task model is trained in a self-training pipeline where the knowledge in rules set is consolidated with self-learned high-confidence rules. Finally, for the low-confidence rules, KICE solicits human-enlightened understanding and expands the knowledge coverage for better task model training. Our framework is verified on relation extraction (RE) task, and the experiments on TACRED show that the model performance (F1) grows from 33.24% to 79.84% with the enrichment of knowledge, outperforming all the baselines including other knowledgeable methods.

lbeit having gained significant progress lately, large-scale graph representation learning remains expensive to train and deploy for two main reasons: (i) the repetitive computation of multi-hop message passing and non-linearity in graph neural networks (GNNs); (ii) the computational cost of complex pairwise contrastive learning loss. Two main contributions are made in this paper targeting this twofold challenge: we first propose an adaptive-view graph neural encoder (AVGE) with a limited number of message passing to accelerate the forward pass computation, and then we propose a structure-aware group discrimination (SAGD) loss in our framework which avoids inefficient pairwise loss computing in most common GCL and improves the performance of the simple group discrimination. By the framework proposed, we manage to bring down the training and inference cost on various large-scale datasets by a significant margin (250x faster inference time) without loss of the downstream-task performance.

Prompt tuning, a recently emerging paradigm, enables the powerful vision-language pre-training models to adapt to downstream tasks in a parameter- and data- efficient way, by learning the "soft prompts" to condition frozen pretraining models. Though effective, it is particularly problematic in the few-shot scenario, where prompt tuning performance is sensitive to the initialization and requires a time-consuming process to find a good initialization, thus restricting the fast adaptation ability of the pre-training models. In addition, prompt tuning could undermine the generalizability of the pre-training models, because the learnable prompt tokens are easy to overfit to the limited training samples. To address these issues, we introduce a novel Gradient-RegulAted Meta-prompt learning (GRAM) framework that jointly meta-learns an efficient soft prompt initialization for better adaptation and a lightweight gradient regulating function for strong cross-domain generalizability in a meta-learning paradigm using only the unlabeled image-text pre-training data. Rather than designing a specific prompt tuning method, our GRAM can be easily incorporated into various prompt tuning methods in a model-agnostic way, and comprehensive experiments show that GRAM brings about consistent improvement for them in several settings (i.e., few-shot learning, cross-domain generalization, cross-dataset generalization, etc.) over 11 datasets. Further, experiments show that GRAM enables the orthogonal methods of textual and visual prompt tuning to work in a mutually-enhanced way, offering better generalizability beyond the uni-modal prompt tuning methods.

Scene Graph Generation (SGG) aims to extractrelationships in images for vision understanding. Although recent works have made steady progress on SGG, they still suffer long-tail distribution issues that tail-predicates are more costly to train and hard to distinguish due to a small amount of annotated data compared to frequent predicates. Existing re-balancing strategies try to handle it via prior rules but are still confined to pre-defined conditions, which are not scalable for various models and datasets. In this paper, we propose a Cross-modal prediCate boosting (CaCao) framework, where a visually-prompted language model is learned to generate diverse fine-grained predicates in a low-resource way. The proposed CaCao can be applied in a plug-and-play fashion and automatically strengthen existing SGG to tackle the long-tailed problem. Based on that, we further introduce a novel Entangled cross-modal prompt approach for open-world predicate scene graph generation (Epic), where models can generalize to unseen predicates in a zero-shot manner. Comprehensive experiments on three benchmark datasets show that CaCao consistently boosts the performance of multiple scene graph generation models in a model-agnostic way. Moreover, our Epic achieves competitive performance on open-world predicate prediction. The data and code for this paper are publicly available.1

Conventional multi-label classification (MLC) methods assume that all samples are fully labeled and identically distributed. Unfortunately, this assumption is unrealistic in large-scale MLC data that has long-tailed (LT) distribution and partial labels (PL). To address the problem, we introduce a novel task, Partial labeling and Long-Tailed Multi-Label Classification (PLT-MLC), to jointly consider the above two imperfect learning environments. Not surprisingly, we find that most LT-MLC and PL-MLC approaches fail to solve the PLT-MLC, resulting in significant performance degradation on the two proposed PLT-MLC benchmarks. Therefore, we propose an end-to-end learning framework: COrrection → ModificatIon → balanCe, abbreviated as COMIC. Our bootstrapping philosophy is to simultaneously correct the missing labels (Correction) with convinced prediction confidence over a class-aware threshold and to learn from these recall labels during training. We next propose a novel multi-focal modifier loss that simultaneously addresses head-tail imbalance and positive-negative imbalance to adaptively modify the attention to different samples (Modification) under the LT class distribution. In addition, we develop a balanced training strategy by distilling the model’s learning effect from head and tail samples, and thus design a balanced classifier (Balance) conditioned on the head and tail learning effect to maintain stable performance for all samples. Our experimental study shows that the proposed COMIC significantly outperforms general MLC, LT-MLC and PL-MLC methods in terms of effectiveness and robustness on our newly created PLT-MLC datasets. Codes and benchmarks are available on the link https://https://github.com/wannature/COMIC

Large-scale multi-modal contrastive learning frameworks like CLIP typically require a large amount of image-text samples for training. However, these samples are always collected continuously in real scenarios. This paper discusses the feasibility of continual CLIP training using streaming data. Unlike continual learning based on self-supervised learning methods for pure images, which is empirically robust against catastrophic forgetting, CLIP’s performance degeneration in the continual setting is significant and non-neglectable. By analyzing the changes in the model’s representation space during continual CLIP training from a spatial geometry perspective, we explore and summarize these spatial variations as Spatial Disorder (SD), which can be divided into Intra-modal Rotation and Inter-modal Deviation. Moreover, we empirically and theoretically demonstrate how SD leads to a performance decline for CLIP on cross-modal retrieval tasks. To alleviate SD, we propose a new continual vision-language representation learning framework Mod-X: Maintain off-diagonal information-matriX. By selectively aligning the off-diagonal information distribution of contrastive matrices, the Mod-X improves the capability of the multi-modal model by maintaining the multi-modal representation space alignment on the old data domain during continuously fitting the new training data domain. Experiments on commonly used datasets with different scales and scopes have demonstrated the effectiveness of our method.

Dynamic early exiting has been proven to improve the inference speed of the pre-trained language model like BERT. However, all samples must go through all consecutive layers before early exiting and more complex samples usually go through more layers, which still exists redundant computation. In this paper, we propose a novel dynamic early exiting combined with layer skipping for BERT inference named SmartBERT, which adds a skipping gate and an exiting operator into each layer of BERT. SmartBERT can adaptively skip some layers and adaptively choose whether to exit. Besides, we propose cross-layer contrastive learning and combine it into our training phases to boost the intermediate layers and classifiers which would be beneficial for early exiting. To keep the inconsistent usage of skipping gates between training and inference phases, we propose a hard weight mechanism during training phase. We conduct experiments on eight classification datasets of the GLUE benchmark. Experimental results show that SmartBERT achieves 2-3× computation reduction with minimal accuracy drops compared with BERT and our method outperforms previous methods in both efficiency and accuracy. Moreover, in some complex datasets, we prove that the early exiting based on entropy hardly works, and the skipping mechanism is essential for reducing computation.

Federated learning aims to train a better global model without sharing the sensitive training samples (usually images) of local clients. Since the sample distributions in local clients tend to be different from each other (i.e., non-IID), one of the major challenges for federated learning is to alleviate model degradation when aggregating local models. The degradation can be attributed to the weight divergence that quantifies the difference of local models from different training processes. Furthermore, non-IID also results in feature space heterogeneity during local training, making neurons of local models in the same location have different functions and further exacerbating weight divergence. In this paper, we demonstrate that the problem can be solved by sharing information from the non-sensitive modality (e.g., metadata, non-sensitive descriptions, etc.) while keeping the sensitive information of images protected. In particular, we propose Federated Learning with Adversarial Example and Adversarial Identifier (FedAA) that trains adversarial examples based on the shared non-sensitive modality to fine-tune local models before global aggregation. The training of local models is enhanced by client identifiers that discriminate the source of inputs to force different local models to get similar outputs and be more homogeneous during the local training. Experiments show that FedAA significantly outperforms recent non-IID federated learning algorithms while preserving image privac, by sharing information from non-sensitive modalities.

Owing to the unrestricted nature of the content in the training data, large text-to-image diffusion models, such as Stable Diffusion (SD), are capable of generating images with potentially copyrighted or dangerous content based on corresponding textual concepts information. This includes specific intellectual property (IP), human faces, and various artistic styles. However, Negative Prompt, a widely used method for content removal, frequently fails to conceal this content due to inherent limitations in its inference logic. In this work, we propose a novel strategy named Degeneration-Tuning (DT) to shield contents of unwanted concepts from SD weights. By utilizing Scrambled Grid to reconstruct the correlation between undesired concepts and their corresponding image domain, we guide SD to generate meaningless content when such textual concepts are provided as input. As this adaptation occurs at the level of the model's weights, the SD, after DT, can be grafted onto other conditional diffusion frameworks like ControlNet to shield unwanted concepts. In addition to qualitatively showcasing the effectiveness of our DT method in protecting various types of concepts, a quantitative comparison of the SD before and after DT indicates that the DT method does not significantly impact the generative quality of other contents. The FID and IS scores of the model on COCO-30K exhibit only minor changes after DT, shifting from 12.61 and 39.20 to 13.04 and 38.25, respectively, which clearly outperforms the previous methods.

Temporal grounding is the task of locating a specific segment from an untrimmed video according to a query sentence. This task has achieved significant momentum in the computer vision community as it enables activity grounding beyond pre-defined activity classes by utilizing the semantic diversity of natural language descriptions. The semantic diversity is rooted in the principle of compositionality in linguistics, where novel semantics can be systematically described by combining known words in novel ways (compositional generalization). However, existing temporal grounding datasets are not carefully designed to evaluate the compositional generalizability. To systematically benchmark the compositional generalizability of temporal grounding models, we introduce a new Compositional Temporal Grounding task and construct two new dataset splits, i.e., Charades-CG and ActivityNet-CG. We empirically find that they fail to generalize to queries with novel combinations of seen words. We argue that the inherent compositional structure (i.e., composition constituents and their relationships) inside the videos and language is the crucial factor to achieve compositional generalization. Based on this insight, we propose a variational cross-graph reasoning framework that explicitly decomposes video and language into hierarchical semantic graphs, respectively, and learns fine-grained semantic correspondence between the two graphs. Meanwhile, we introduce a novel adaptive structured semantics learning approach to derive the structure-informed and domain-generalizable graph representations, which facilitate the fine-grained semantic correspondence reasoning between the two graphs. To further evaluate the understanding of the compositional structure, we also introduce a more challenging setting, where one of the components in the novel composition is unseen. This requires more sophisticated understanding of the compositional structure to infer the potential semantics of the unseen word based on the other learned composition constituents appearing in both the video and language context, and their relationships. Extensive experiments validate the superior compositional generalizability of our approach, demonstrating its ability to handle queries with novel combinations of seen words as well as novel words in the testing composition.

Data augmentation has become one of the keys to alleviating the over-fitting of models on training data and improving the generalization capabilities on testing data. Most existing data augmentation methods only focus on one modality, which is incapable when facing multiple data modalities. Some prior works try to interpolate with random coefficients in the latent space to generate new samples, which can generically work for any data modality. However, these works ignore the extra information conveyed by multimodality data. In fact, the extra information in one modality can provide semantic directions to generate more meaningful samples in another modality. This paper proposes Cross-modal Data Augmentation (CMDA), a simple yet effective data augmentation method to alleviate the over-fitting issue and improve the generalization performance. We evaluate CMDA on unsupervised and supervised tasks of different modalities, on which CMDA consistently and significantly outperforms baselines. For instance, CMDA improves the unsupervised anomaly detection baseline in vision modality from the AUROC$76.46\%, 73.07\%$and 64.36% to$83.25\%, 76.22\%$and 70.57% on three different datasets, respectively. Besides, extensive experiments demonstrate that CMDA is applicable to various neural network architectures. Furthermore, prior methods that interpolate in the latent space need to work with downstream tasks to construct the latent space. In contrast, CMDA can work with or without downstream tasks, which makes the applicability of CMDA more extensive. The source code is publicly available for non-commercial or research use athttps://github.com/Anfeather/CMDA

It is a consensus that small models perform quite poorly under the paradigm of self-supervised contrastive learning. Existing methods usually adopt a large off-the-shelf model to transfer knowledge to the small one via distillation. Despite their effectiveness, distillation-based methods may not be suitable for some resource-restricted scenarios due to the huge computational expenses of deploying a large model. In this paper, we study the issue of training self-supervised small models without distillation signals. We first evaluate the representation spaces of the small models and make two non-negligible observations: (i) the small models can complete the pretext task without overfitting despite their limited capacity and (ii) they universally suffer the problem of over clustering. Then we verify multiple assumptions that are considered to alleviate the over-clustering phenomenon. Finally, we combine the validated techniques and improve the baseline performances of five small architectures with considerable margins, which indicates that training small self-supervised contrastive models is feasible even without distillation signals. The code is available at https://github.com/WOWNICE/ssl-small.

Machine Reading Comprehension (MRC) reveals the ability to understand a given text passage and answer questions based on it.Existing research works in MRC rely heavily on large-size models and corpus to improve the performance evaluated by metrics such as Exact Match (EM ) and F 1 .However, such a paradigm lacks sufficient interpretation to model capability and can not efficiently train a model with a large corpus.In this paper, we argue that a deep understanding of model capabilities and data properties can help us feed a model with appropriate training data based on its learning status.Specifically, we design an MRC capability assessment framework that assesses model capabilities in an explainable and multi-dimensional manner.Based on it, we further uncover and disentangle the connections between various data properties and model performance.Finally, to verify the effectiveness of the proposed MRC capability assessment framework, we incorporate it into a curriculum learning pipeline and devise a Capability Boundary Breakthrough Curriculum (CBBC) strategy, which performs a model capability-based training to maximize the data value and improve training efficiency.Extensive experiments demonstrate that our approach significantly improves performance, achieving up to an 11.22% / 8.71% improvement of EM / F 1 on MRC tasks.

Temporal grounding in videos aims to localize one target video segment that semantically corresponds to a given query sentence. Thanks to the semantic diversity of natural language descriptions, temporal grounding allows activity grounding beyond pre-defined classes and has received increasing attention in recent years. The semantic diversity is rooted in the principle of compositionality in linguistics, where novel semantics can be systematically described by combining known words in novel ways (compositional generalization). However, current temporal grounding datasets do not specifically test for the compositional generalizability. To systematically measure the compositional generalizability of temporal grounding models, we introduce a new Compositional Temporal Grounding task and construct two new dataset splits, i.e., Charades-CG and ActivityNet-CG. Evaluating the state-of-the-art methods on our new dataset splits, we empirically find that they fail to generalize to queries with novel combinations of seen words. To tackle this challenge, we propose a variational cross-graph reasoning framework that explicitly decomposes video and language into multiple structured hierarchies and learns fine-grained semantic correspondence among them. Experiments illustrate the superior compositional generalizability of our approach. The repository of this work is at ht tps: / / gi thub. com/YYJMJC/ Composi tional- Temporal-Grounding.

Meta-learning enables models to adapt to new environments rapidly with a few training examples. Current gradient-based meta-learning methods concentrate on finding good model-agnostic initialization (meta-weights) for learners. In this paper, we aim to obtain better meta-learners by co-optimizing the architecture and meta-weights simultaneously. Existing NAS-based meta-learning methods apply a two-stage strategy, i.e., first searching architectures and then re-training meta-weights on the searched architecture. However, this two-stage strategy would break the mutual impact of the architecture and meta-weights since they are optimized separately. Differently, we propose progressive connection consolidation, fixing the architecture layer by layer, in which the layer with the largest weight value would be fixed first. In this way, we can jointly search architectures and train the meta-weights on fixed layers. Besides, to improve the generalization performance of the searched meta-learner on all tasks, we propose a more effective rule for co-optimization, namely Connection-Adaptive Meta-learning (CAML). By searching only once, we can obtain both adaptive architecture and meta-weights for meta-learning. Extensive experiments show that our method achieves state-of-the-art performance with 3x less computational cost, revealing our method's effectiveness and efficiency.

Novel category discovery aims at adapting models trained on known categories to novel categories.Previous works only focus on the scenario where known and novel categories are of the same granularity.In this paper, we investigate a new practical scenario called Fine-grained Category Discovery under Coarsegrained supervision (FCDC).FCDC aims at discovering fine-grained categories with only coarse-grained labeled data, which can adapt models to categories of different granularity from known ones and reduce significant labeling cost.It is also a challenging task since supervised training on coarse-grained categories tends to focus on inter-class distance (distance between coarse-grained classes) but ignore intra-class distance (distance between fine-grained sub-classes) which is essential for separating fine-grained categories.Considering most current methods cannot transfer knowledge from coarse-grained level to fine-grained level, we propose a hierarchical weighted self-contrastive network by building a novel weighted self-contrastive module and combining it with supervised learning in a hierarchical manner.Extensive experiments on public datasets show both effectiveness and efficiency of our model over compared methods.Code and data are available at https://github.com/Lackel/ Hierarchical_Weighted_SCL.

Text error correction aims to correct the errors in text sequences such as those typed by humans or generated by speech recognition models.Previous error correction methods usually take the source (incorrect) sentence as encoder input and generate the target (correct) sentence through the decoder.Since the error rate of the incorrect sentence is usually low (e.g., 10%), the correction model can only learn to correct on limited error tokens but trivially copy on most tokens (correct tokens), which harms the effective training of error correction.In this paper, we argue that the correct tokens should be better utilized to facilitate effective training and then propose a simple yet effective masking strategy to achieve this goal.Specifically, we randomly mask out a part of the correct tokens in the source sentence and let the model learn to not only correct the original error tokens but also predict the masked tokens based on their context information.Our method enjoys several advantages: 1) it alleviates trivial copy; 2) it leverages effective training signals from correct tokens; 3) it is a plug-and-play module and can be applied to different models and tasks.Experiments on spelling error correction and speech recognition error correction on Mandarin datasets and grammar error correction on English datasets with both autoregressive and non-autoregressive generation models show that our method improves the correction accuracy consistently 1 .

Existing metrics for assessing question generation not only require costly human reference but also fail to take into account the input context of generation, rendering the lack of deep understanding of the relevance between the generated questions and input contexts.As a result, they may wrongly penalize a legitimate and reasonable candidate question when it (i) involves complicated reasoning with the context or (ii) can be grounded by multiple evidences in the context.In this paper, we propose QRelScore, a context-aware Relevance evaluation metric for Question Generation.Based on off-the-shelf language models such as BERT and GPT2, QRelScore employs both word-level hierarchical matching and sentence-level prompt-based generation to cope with the complicated reasoning and diverse generation from multiple evidences, respectively.Compared with existing metrics, our experiments demonstrate that QRelScore is able to achieve a higher correlation with human judgments while being much more robust to adversarial samples.

Recent years have seen a surge of interest in meta-learning techniques for tackling the few-shot learning (FSL) problem. However, the meta-learner is prone to overfitting since there are only a few available samples, which can be identified as sampling noise on a clean dataset. Besides, when handling the data with noisy labels, the meta-learner could be extremely sensitive to label noise on a corrupted dataset. To address these two challenges, we present Eigen-Reptile (ER) that updates the meta-parameters with the main direction of historical task-specific parameters. Specifically, the main direction is computed in a fast way, where the scale of the calculated matrix is related to the number of gradient steps for the specific task instead of the number of parameters. Furthermore, to obtain a more accurate main direction for Eigen-Reptile in the presence of many noisy labels, we further propose Introspective Self-paced Learning (ISPL). We have theoretically and experimentally demonstrated the soundness and effectiveness of the proposed Eigen-Reptile and ISPL. Particularly, our experiments on different tasks show that the proposed method is able to outperform or achieve highly competitive performance compared with other gradient-based methods with or without noisy labels. The code and data for the proposed method are provided for research purposes https://github.com/Anfeather/Eigen-Reptile.

Graph contrastive learning has gained significant progress recently. However, existing works have rarely explored non-aligned node-node contrasting. In this paper, we propose a novel graph contrastive learning method named RoSA that focuses on utilizing non-aligned augmented views for node-level representation learning. First, we leverage the earth mover's distance to model the minimum effort to transform the distribution of one view to the other as our contrastive objective, which does not require alignment between views. Then we introduce adversarial training as an auxiliary method to increase sampling diversity and enhance the robustness of our model. Experimental results show that RoSA outperforms a series of graph contrastive learning frameworks on homophilous, non-homophilous and dynamic graphs, which validates the effectiveness of our work. To the best of our awareness, RoSA is the first work focuses on the non-aligned node-node graph contrastive learning problem. Our codes are available at: https://github.com/ZhuYun97/RoSA

While annotating decent amounts of data to satisfy sophisticated learning models can be cost-prohibitive for many real-world applications. Active learning (AL) and semi-supervised learning (SSL) are two effective, but often isolated, means to alleviate the data-hungry problem. Some recent studies explored the potential of combining AL and SSL to better probe the unlabeled data. However, almost all these contemporary SSL-AL works use a simple combination strategy, ignoring SSL and AL's inherent relation. Further, other methods suffer from high computational costs when dealing with large-scale, high-dimensional datasets. Motivated by the industry practice of labeling data, we propose an innovative Inconsistency-based virtual aDvErsarial Active Learning (IDEAL) algorithm to further investigate SSL-AL's potential superiority and achieve mutual enhancement of AL and SSL, i.e., SSL propagates label information to unlabeled samples and provides smoothed embeddings for AL, while AL excludes samples with inconsistent predictions and considerable uncertainty for SSL. We estimate unlabeled samples' inconsistency by augmentation strategies of different granularities, including fine-grained continuous perturbation exploration and coarse-grained data transformations. Extensive experiments, in both text and image domains, validate the effectiveness of the proposed algorithm, comparing it against state-of-the-art baselines. Two real-world case studies visualize the practical industrial value of applying and deploying the proposed data sampling algorithm.

Understanding human emotions is a crucial ability for intelligent robots to provide better human-robot interactions. The existing works are limited to trimmed video-level emotion classification, failing to locate the temporal window corresponding to the emotion. In this paper, we introduce a new task, named Temporal Emotion Localization in videos (TEL), which aims to detect human emotions and localize their corresponding temporal boundaries in untrimmed videos with aligned subtitles. TEL presents three unique challenges compared to temporal action localization: 1) The emotions have extremely varied temporal dynamics; 2) The emotion cues are embedded in both appearances and complex plots; 3) The fine-grained temporal annotations are complicated and labor-intensive. To address the first two challenges, we propose a novel dilated context integrated network with a coarse-fine two-stream architecture. The coarse stream captures varied temporal dynamics by modeling multi-granularity temporal contexts. The fine stream achieves complex plots understanding by reasoning the dependency between the multi-granularity temporal contexts from the coarse stream and adaptively integrates them into fine-grained video segment features. To address the third challenge, we introduce a cross-modal consensus learning paradigm, which leverages the inherent semantic consensus between the aligned video and subtitle to achieve weakly-supervised learning. We contribute a new testing set with 3,000 manually-annotated temporal boundaries so that future research on the TEL problem can be quantitatively evaluated. Extensive experiments show the effectiveness of our approach on temporal emotion localization. The repository of this work is at https://github.com/YYJMJC/TemporalEmotion-Localization-in-Videos.

Multimedia recommendation aims to predict user preferences where users interact with multimodal items. Collaborative filtering based on graph convolutional networks manifests impressive performance gains in multimedia recommendation. This is attributed to the capability of learning good user and item embeddings by aggregating the collaborative signals from high-order neighbors. However, previous researches [37,38] fail to explicitly mine different behavior patterns (i.e., item categories, common user interests) by exploiting user-item and item-item graphs simultaneously, which plays an important role in modeling user preferences. And it is the lack of different behavior pattern constraints and multimodal feature reconciliations that results in performance degradation. Towards this end, We propose a Hybrid Clustering Graph Convolutional Network (HCGCN) for multimedia recommendation. We perform high-order graph convolutions inside user-item clusters and item-item clusters to capture various user behavior patterns. Meanwhile, we design corresponding clustering losses to enhance user-item preference feedback and multimodal representation learning constraint to adjust the modality importance, making more accurate recommendations. Experimental results on three real-world multimedia datasets not only demonstrate the significant improvement of our model over the state-of-the-art methods, but also validate the effectiveness of integrating hybrid user behavior patterns for multimedia recommendation.

Large-scale vision-language pre-training has shown impressive advances in a wide range of downstream tasks. Existing methods mainly model the cross-modal alignment by the similarity of the global representations of images and text, or advanced cross-modal attention upon image and text features. However, they fail to explicitly learn the fine-grained semantic alignment between visual regions and textual phrases, as only global image-text alignment information is available. In this paper, we introduce LOUPE, a fine-grained semantically aLigned visiOn-langUage PrE-training framework, which learns fine-grained semantic alignment from the novel perspective of game-theoretic interactions. To efficiently estimate the game-theoretic interactions, we further propose an uncertainty-aware neural Shapley interaction learning module. Experiments show that LOUPE achieves state-of-the-art performance on a variety of vision-language tasks. Without any object-level human annotations and fine-tuning, LOUPE achieves competitive performance on object detection and visual grounding. More importantly, LOUPE opens a new promising direction of learning fine-grained semantics from large-scale raw image-text pairs.

Training deep models for RGB-D salient object detection (SOD) often requires a large number of labeled RGB-D images. However, RGB-D data is not easily acquired, which limits the development of RGB-D SOD techniques. To alleviate this issue, we present a Dual-Semi RGB-D Salient Object Detection Network (DS-Net) to leverage unlabeled RGB images for boosting RGB-D saliency detection. We first devise a depth decoupling convolutional neural network (DDCNN), which contains a depth estimation branch and a saliency detection branch. The depth estimation branch is trained with RGB-D images and then used to estimate the pseudo depth maps for all unlabeled RGB images to form the paired data. The saliency detection branch is used to fuse the RGB feature and depth feature to predict the RGB-D saliency. Then, the whole DDCNN is assigned as the backbone in a teacher-student framework for semi-supervised learning. Moreover, we also introduce a consistency loss on the intermediate attention and saliency maps for the unlabeled data, as well as a supervised depth and saliency loss for labeled data. Experimental results on seven widely-used benchmark datasets demonstrate that our DDCNN outperforms state-of-the-art methods both quantitatively and qualitatively. We also demonstrate that our semi-supervised DS-Net can further improve the performance, even when using an RGB image with the pseudo depth map.

With recent advances in distantly supervised (DS) relation extraction (RE), considerable attention is attracted to leverage multi-instance learning (MIL) to distill high-quality supervision from the noisy DS. Here, we go beyond label noise and identify the key bottleneck of DS-MIL to be its low data utilization: as high-quality supervision being refined by MIL, MIL abandons a large amount of training instances, which leads to a low data utilization and hinders model training from having abundant supervision. In this paper, we propose collaborative adversarial training to improve the data utilization, which coordinates virtual adversarial training (VAT) and adversarial training (AT) at different levels. Specifically, since VAT is label-free, we employ the instance-level VAT to recycle instances abandoned by MIL. Besides, we deploy AT at the bag-level to unleash the full potential of the high-quality supervision got by MIL. Our proposed method brings consistent improvements (∼ 5 absolute AUC score) to the previous state of the art, which verifies the importance of the data utilization issue and the effectiveness of our method.

In this paper, we propose a novel graph learning framework for phrase grounding in the image. Developing from the sequential to the dense graph model, existing works capture coarse-grained context but fail to distinguish the diversity of context among phrases and image regions. In contrast, we pay special attention to different motifs implied in the context of the scene graph and devise the disentangled graph network to integrate the motif-aware contextual information into representations. Besides, we adopt interventional strategies at the feature and the structure levels to consolidate and generalize representations. Finally, the cross-modal attention network is utilized to fuse intra-modal features, where each phrase can be computed similarity with regions to select the best-grounded one. We validate the efficiency of disentangled and interventional graph network (DIGN) through a series of ablation studies, and our model achieves state-of-the-art performance on Flickr30K Entities and ReferIt Game benchmarks.

The contemporary visual captioning models frequently hallucinate objects that are not actually in a scene, due to the visual misclassification or over-reliance on priors that resulting in the semantic inconsistency between the visual information and the target lexical words. The most common way is to encourage the captioning model to dynamically link generated object words or phrases to appropriate regions of the image, i.e., the grounded image captioning (GIC). However, GIC utilizes an auxiliary task (grounding objects) that has not solved the key issue of object hallucination, i.e., the semantic inconsistency. In this paper, we take a novel perspective on the issue above: exploiting the semantic coherency between the visual and language modalities. Specifically, we propose the Consensus Rraph Representation Learning framework (CGRL) for GIC that incorporates a consensus representation into the grounded captioning pipeline. The consensus is learned by aligning the visual graph (e.g., scene graph) to the language graph that consider both the nodes and edges in a graph. With the aligned consensus, the captioning model can capture both the correct linguistic characteristics and visual relevance, and then grounding appropriate image regions further. We validate the effectiveness of our model, with a significant decline in object hallucination (-9% CHAIRi) on the Flickr30k Entities dataset. Besides, our CGRL also evaluated by several automatic metrics and human evaluation, the results indicate that the proposed approach can simultaneously improve the performance of image captioning (+2.9 Cider) and grounding (+2.3 F1LOC}).

Tao Chen, Haizhou Shi, Siliang Tang, Zhigang Chen, Fei Wu, Yueting Zhuang. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

The performance of computer vision models significantly improves with more labeled data. However, the acquisition of labeled data is limited by the high cost. To mitigate the reliance on large labeled datasets, active learning (AL) and semi-supervised learning (SSL) are frequently adopted. Although current mainstream methods begin to combine SSL and AL (SSL-AL) to excavate the diverse expressions of unlabeled samples, these methods’ fully supervised task models are still trained only with labeled data. Besides, these method’s SSL-AL frameworks suffer from mismatch problems. Here, we propose a graph-based SSL-AL framework to unleash the SSL task models’ power and make an effective SSL-AL interaction. In the framework, SSL leverages graph-based label propagation to deliver virtual labels to unlabeled samples, rendering AL samples’ structural distribution and boosting AL. AL finds samples near the clusters’ boundary to help SSL perform better label propagation by exploiting adversarial examples. The information exchange in the closed-loop realizes mutual enhancement of SSL and AL. Experimental results show that our method outperforms the state-of-the-art methods against classification and segmentation benchmarks.

Video-and-Language Inference is a recently proposed task for joint video-and-language understanding. This new task requires a model to draw inference on whether a natural language statement entails or contradicts a given video clip. In this paper, we study how to address three critical challenges for this task: judging the global correctness of the statement involved multiple semantic meanings, joint reasoning over video and subtitles, and modeling long-range relationships and complex social interactions. First, we propose an adaptive hierarchical graph network that achieves in-depth understanding of the video over complex interactions. Specifically, it performs joint reasoning over video and subtitles in three hierarchies, where the graph structure is adaptively adjusted according to the semantic structures of the statement. Secondly, we introduce semantic coherence learning to explicitly encourage the semantic coherence of the adaptive hierarchical graph network from three hierarchies. The semantic coherence learning can further improve the alignment between vision and linguistics, and the coherence across a sequence of video segments. Experimental results show that our method significantly outperforms the baseline by a large margin.

The recently emerged weakly-supervised object localization (WSOL) methods can learn to localize an object in the image only using image-level labels. Previous works endeavor to perceive the interval objects from the small and sparse discriminative attention map, yet ignoring the co-occurrence confounder (e.g., duck and water), which makes the model inspection (e.g., CAM) hard to distinguish between the object and context. In this paper, we make an early attempt to tackle this challenge via causal intervention (CI). Our proposed method, dubbed CI-CAM, explores the causalities among image features, contexts, and categories to eliminate the biased object-context entanglement in the class activation maps thus improving the accuracy of object localization. Extensive experiments on several benchmarks demonstrate the effectiveness of CI-CAM in learning the clear object boundary from confounding contexts. Particularly, on the CUB-200-2011 which severely suffers from the co-occurrence confounder, CI-CAM significantly outperforms the traditional CAM-based baseline (58.39% vs 52.4% in Top-1 localization accuracy). While in more general scenarios such as ILSVRC 2016, CI-CAM can also perform on par with the state of the arts.

The task of visual question generation (VQG) aims to generate human-like neural questions from an image and potentially other side information (e.g., answer type or the answer itself). Existing works often suffer from the severe one image to many questions mapping problem, which generates uninformative and non-referential questions. Recent work has demonstrated that by leveraging double visual and answer hints, a model can faithfully generate much better quality questions. However, visual hints are not available naturally. Despite they proposed a simple rule-based similarity matching method to obtain candidate visual hints, they could be very noisy practically and thus restrict the quality of generated questions. In this paper, we present a novel learning approach for double-hints based VQG, which can be cast as a weakly supervised learning problem with noises. The key rationale is that the salient visual regions of interest can be viewed as a constraint to improve the generation procedure for producing high-quality questions. As a result, given the predicted salient visual regions of interest, we can focus on estimating the probability of being ground-truth questions, which in turn implicitly measures the quality of predicted visual hints. Experimental results on two benchmark datasets show that our proposed method outperforms the state-of-the-art approaches by a large margin on a variety of metrics, including both automatic machine metrics and human evaluation.

长期以来困扰人工智能领域的一个问题是: 人工智能是否具有创造力, 或者说, 算法的推理过程是否可以具有创造性. 本文从思维科学的角度探讨人工智能创造力的问题. 首先, 列举形象思维推理的相关研究; 然后, 重点介绍一种特殊的视觉知识表示形式, 即视觉场景图; 最后, 详细介绍视觉场景图构造问题与潜在应用. 所有证据表明, 视觉知识和视觉思维不仅可以改善当前人工智能任务的性能, 而且可以用于机器创造力的实践.

Electronic medical records (EMRs) play an important role in medical data mining and sequential data learning. In this article, we propose to use a sequential neural network with dynamic content-based memories to predict future medications, given EMRs. The local-global memory neural network contains two layers of memories: the local memory and the global memory. Particularly, our method learns the hidden knowledge within EMRs by locally remembering individual patterns of a patient (via local memory) and globally remembering group evidence of disease (via global memory). In addition, we show how our model can be modified to classify the hidden states of EMRs from different patients at each time step into different phases that indicate the progressions of medications in terms of a specific disease, in an unsupervised manner. Experimental results on real EMRs data sets show that, by learning EMRs with external local and global memories, with regard to a given disease, our model improves the prediction performance compared with several alternative methods.

In this paper, we focus on the task query-based video localization, i.e., localizing a query in a long and untrimmed video. The prevailing solutions for this problem can be grouped into two categories: i) Top-down approach: It pre-cuts the video into a set of moment candidates, then it does classification and regression for each candidate; ii) Bottom-up approach: It injects the whole query content into each video frame, then it predicts the probabilities of each frame as a ground truth segment boundary (i.e., start or end). Both two frameworks have respective shortcomings: the top-down models suffer from heavy computations and they are sensitive to the heuristic rules, while the performance of bottom-up models is behind the performance of top-down counterpart thus far. However, we argue that the performance of bottom-up framework is severely underestimated by current unreasonable designs, including both the backbone and head network. To this end, we design a novel bottom-up model: Graph-FPN with Dense Predictions (GDP). For the backbone, GDP firstly generates a frame feature pyramid to capture multi-level semantics, then it utilizes graph convolution to encode the plentiful scene relationships, which incidentally mitigates the semantic gaps in the multi-scale feature pyramid. For the head network, GDP regards all frames falling in the ground truth segment as the foreground, and each foreground frame regresses the unique distances from its location to bi-directional boundaries. Extensive experiments on two challenging query-based video localization tasks (natural language video localization and video relocalization), involving four challenging benchmarks (TACoS, Charades-STA, ActivityNet Captions, and Activity-VRL), have shown that GDP surpasses the state-of-the-art top-down models.

Measuring the scholarly impact of a document without citations is an important and challenging problem. Existing approaches such as Document Influence Model (DIM) are based on dynamic topic models, which only consider the word frequency change. In this paper, we use both frequency changes and word semantic shifts to measure document influence by developing a neural network framework. Our model has three steps. Firstly, we train the word embeddings for different time periods. Subsequently, we propose an unsupervised method to align vectors for different time periods. Finally, we compute the influence value of documents. Our experimental results show that our model outperforms DIM.

Visual navigation is a task of training an embodied agent by intelligently navigating to a target object (e.g., television) using only visual observations. A key challenge for current deep reinforcement learning models lies in the requirements for a large amount of training data. It is exceedingly expensive to construct sufficient 3D synthetic environments annotated with the target object information. In this paper, we focus on visual navigation in the low-resource setting, where we have only a few training environments annotated with object information. We propose a novel unsupervised reinforcement learning approach to learn transferable meta-skills (e.g., bypass obstacles, go straight) from unannotated environments without any supervisory signals. The agent can then fast adapt to visual navigation through learning a high-level master policy to combine these meta-skills, when the visual-navigation-specified reward is provided. Experimental results show that our method significantly outperforms the baseline by 53.34% relatively on SPL, and further qualitative analysis demonstrates that our method learns transferable motor primitives for visual navigation.

Today text classification models have been widely used. However, these classifiers are found to be easily fooled by adversarial examples. Fortunately, standard attacking methods generate adversarial texts in a pair-wise way, that is, an adversarial text can only be created from a real-world text by replacing a few words. In many applications, these texts are limited in numbers, therefore their corresponding adversarial examples are often not diverse enough and sometimes hard to read, thus can be easily detected by humans and cannot create chaos at a large scale. In this paper, we propose an end to end solution to efficiently generate adversarial texts from scratch using generative models, which are not restricted to perturbing the given texts. We call it unrestricted adversarial text generation. Specifically, we train a conditional variational autoencoder (VAE) with an additional adversarial loss to guide the generation of adversarial examples. Moreover, to improve the validity of adversarial texts, we utilize discrimators and the training framework of generative adversarial networks (GANs) to make adversarial texts consistent with real data. Experimental results on sentiment analysis demonstrate the scalability and efficiency of our method. It can attack text classification models with a higher success rate than existing methods, and provide acceptable quality for humans in the meantime.

The task of Grounded Video Description~(GVD) is to generate sentences whose objects can be grounded with the bounding boxes in the video frames. Existing works often fail to exploit structural information both in modeling the relationships among the region proposals and in attending them for text generation. To address these issues, we cast the GVD task as a spatial-temporal Graph-to-Sequence learning problem, where we model video frames as spatial-temporal sequence graph in order to better capture implicit structural relationships. In particular, we exploit two ways to construct a sequence graph that captures spatial-temporal correlations among different objects in each frame and further present a novel graph topology refinement technique to discover optimal underlying graph structure. In addition, we also present hierarchical attention mechanism to attend sequence graph in different resolution levels for better generating the sentences. Our extensive experiments demonstrate the effectiveness of our proposed method compared to state-of-the-art methods.

The recent success of Distant Supervision (DS) brings abundant labeled data for the task of fine-grained entity typing (FET) without human annotation. However, the heuristically generated labels inevitably bring a significant distribution gap, namely dataset shift, between the distantly labeled training set and the manually curated test set. Considerable efforts have been made to alleviate this problem from the label perspective by either intelligently denoising the training labels, or designing noise-aware loss functions. Despite their progress, the dataset shift can hardly be eliminated completely. In this work, complementary to the label perspective, we reconsider this problem from the model perspective: Can we learn a more robust typing model with the existence of dataset shift? To this end, we propose a novel regularization module based on virtual adversarial training (VAT). The proposed approach first uses a self-paced sample selection function to select suitable samples for VAT, then constructs virtual adversarial perturbations based on the selected samples, and finally regularizes the model to be robust to such perturbations. Experiments on two benchmarks demonstrate the effectiveness of the proposed method, with an average 3.8%, 2.5%, and 3.2% improvement in accuracy, Macro F1 and Micro F1 respectively compared to the next best method.

Visual Storytelling~(VIST) is a task to tell a narrative story about a certain topic according to the given photo stream. The existing studies focus on designing complex models, which rely on a huge amount of human-annotated data. However, the annotation of VIST is extremely costly and many topics cannot be covered in the training dataset due to the long-tail topic distribution. In this paper, we focus on enhancing the generalization ability of the VIST model by considering the few-shot setting. Inspired by the way humans tell a story, we propose a topic adaptive storyteller to model the ability of inter-topic generalization. In practice, we apply the gradient-based meta-learning algorithm on multi-modal seq2seq models to endow the model the ability to adapt quickly from topic to topic. Besides, We further propose a prototype encoding structure to model the ability of intra-topic derivation. Specifically, we encode and restore the few training story text to serve as a reference to guide the generation at inference time. Experimental results show that topic adaptation and prototype encoding structure mutually bring benefit to the few-shot model on BLEU and METEOR metric. The further case study shows that the stories generated after few-shot adaptation are more relative and expressive.

Understanding and reasoning over partially observed visual clues are often regarded as a challenging real-world problem even for human beings. In this paper, we present a new visual question answering (VQA) task -- Photo Stream QA, which aims to answer the open-ended questions about a narrative photo stream. Photo Stream QA is more challenging and interesting than the existing VQA tasks, since the temporal and visual variance among photos in the stream is huge and hard to observe. Therefore, instead of learning simple vision-text mappings, the AI algorithms must fill these variance gaps with more recollection, reasoning, even the knowledge from our daily experiences. To tackle the problems in Photo Stream QA, we propose an end-to-end baseline (E-TAA) with a novel Experienced Unit (E-unit) and Three-stage Alternating Attention (TAA). E-unit yields a better visual representation which captures the temporal semantic relation among visual clues in the photo stream, while TAA creates three levels of attention that gradually refines visual features by using the textual representation from the question as the guidance. Experimental results on our developed dataset demonstrate that, as the first attempt at the Photo Stream QA task, E-TAA provides promising results outperforming all the other baseline methods.

Grounded video description (GVD) encourages captioning models to attend to appropriate video regions (e.g., objects) dynamically and generate a description. Such a setting can help explain the decisions of captioning models and prevents the model from hallucinating object words in its description. However, such design mainly focuses on object word generation and thus may ignore fine-grained information and suffer from missing visual concepts. Moreover, relational words (e.g., 'jump left or right') are usual spatio-temporal inference results, i.e., these words cannot be grounded on certain spatial regions. To tackle the above limitations, we design a novel relational graph learning framework for GVD, in which a language-refined scene graph representation is designed to explore fine-grained visual concepts. Furthermore, the refined graph can be regarded as relational inductive knowledge to assist captioning models in selecting the relevant information it needs to generate correct words. We validate the effectiveness of our model through automatic metrics and human evaluation, and the results indicate that our approach can generate more fine-grained and accurate description, and it solves the problem of object hallucination to some extent.

Opinion question machine reading comprehension (MRC) requires a machine to answer questions by analyzing corresponding passages. Compared with traditional MRC tasks where the answer to every question is a segment of text in corresponding passages, opinion question MRC is more challenging because the answer to an opinion question may not appear in corresponding passages but needs to be deduced from multiple sentences. In this study, a novel framework based on neural networks is proposed to address such problems, in which a new hybrid embedding training method combining text features is used. Furthermore, extra attention and output layers which generate auxiliary losses are introduced to jointly train the stacked recurrent neural networks. To deal with imbalance of the dataset, irrelevancy of question and passage is used for data augmentation. Experimental results show that the proposed method achieves state-of-the-art performance. We are the biweekly champion in the opinion question MRC task in Artificial Intelligence Challenger 2018 (AIC2018).

Recently, convolutional neural network (CNN) based models have shown great potential in the task of single image superresolution (SISR). However, many state-of-the-art SISR solutions are reproducing some tricks proven effective in other vision tasks, such as pursuing a deeper model. In this paper, we propose a new solution (named as Multi-Receptive-Field Network - MRFN), which outperforms existing SISR solutions in three different aspects. First, from receptive field: a novel multi-receptive-field (MRF) module is proposed to extract and fuse features in different receptive fields from local to global. Integrating these hierarchical features can generate better mappings on recovering high-fidelity details at different scales. Second, from network architectures: both dense skip connections and deep supervision are utilized to combine features from the current MRF module and preceding ones for training more representative features. Moreover, a deconvolution layer is embedded at the end of the network to avoid artificial priors induced by numerical data pre-processing (e.g., bicubic stretching), and speed up the restoration process. Finally, from error modeling: different from L1 and L2 loss functions, we proposed a novel two-parameter training loss called Weighted Huber loss function which can adaptively adjust the value of back-propagated derivative according to the residual value, thus fit the reconstruction error more effectively. Extensive qualitative and quantitative evaluation results on benchmark datasets demonstrate that our proposed MRFN can achieve more accurate recovering results than most state-of-the-art methods with significantly less complexity.

Vision and language understanding is one of the most fundamental and challenging problems in Multimedia Intelligence. Simultaneously understanding video actions with a related natural language question, and further produces accurate answer is even more challenging since it requires joint modeling information across modality. In the past few years, some studies begin to attack this problem by utilizing attention enhanced deep neural networks. However, simple attention mechanisms such as unidirectional attention fail to yield a better mapping between different modalities. Moreover, none of these Video QA models explore high-level semantics in augmented video-frame level. In this paper, we augmented each frame representation with its context information by a novel feature extractor that combines the advantages of Resnet and a variant of C3D. In addition, we proposed a novel alternating attention network which can alternately attend frame regions, video frames and words in the question in multi-turns. This yields better joint representations of video and question, further help the deep model to discover the deeper relationship between two modalities. Our method outperforms the state-of-the-art Video QA models on two existing video question answering datasets. Further ablation studies proved that our feature extractor and the alternating attention mechanism can improve the performance jointly.

Distant supervision leverages knowledge bases to automatically label instances, thus allowing us to train relation extractor without human annotations. However, the generated training data typically contain massive noise, and may result in poor performances with the vanilla supervised learning. In this paper, we propose to conduct multi-instance learning with a novel Cross-relation Cross-bag Selective Attention (C2SA), which leads to noise-robust training for distant supervised relation extractor. Specifically, we employ the sentence-level selective attention to reduce the effect of noisy or mismatched sentences, while the correlation among relations were captured to improve the quality of attention weights. Moreover, instead of treating all entity-pairs equally, we try to pay more attention to entity-pairs with a higher quality. Similarly, we adopt the selective attention mechanism to achieve this goal. Experiments with two types of relation extractor demonstrate the superiority of the proposed approach over the state-of-the-art, while further ablation studies verify our intuitions and demonstrate the effectiveness of our proposed two techniques.

Xiyuan Yang, Xiaotao Gu, Sheng Lin, Siliang Tang, Yueting Zhuang, Fei Wu, Zhigang Chen, Guoping Hu, Xiang Ren. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Existing methods in the Visual Storytelling field often suffer from the problem of generating general descriptions, while the image contains a lot of meaningful contents remaining unnoticed. The failure of informative story generation can be concluded to the model's incompetence of capturing enough meaningful concepts. The categories of these concepts include entities, attributes, actions, and events, which are in some cases crucial to grounded storytelling. To solve this problem, we propose a method to mine the cross-modal rules to help the model infer these informative concepts given certain visual input. We first build the multimodal transactions by concatenating the CNN activations and the word indices. Then we use the association rule mining algorithm to mine the cross-modal rules, which will be used for the concept inference. With the help of the cross-modal rules, the generated stories are more grounded and informative. Besides, our proposed method holds the advantages of interpretation, expandability, and transferability, indicating potential for wider application. Finally, we leverage these concepts in our encoder-decoder framework with the attention mechanism. We conduct several experiments on the VIsual StoryTelling~(VIST) dataset, the results of which demonstrate the effectiveness of our approach in terms of both automatic metrics and human evaluation. Additional experiments are also conducted showing that our mined cross-modal rules as additional knowledge helps the model gain better performance when trained on a small dataset.

The EmbodiedQA is a task of training an embodied agent by intelligently navigating in a simulated environment and gathering visual information to answer questions. Existing approaches fail to explicitly model the mental imagery function of the agent, while the mental imagery is crucial to embodied cognition, and has a close relation to many high-level meta-skills such as generalization and interpretation. In this paper, we propose a novel Mental Imagery eNhanceD (MIND) module for the embodied agent, as well as a relevant deep reinforcement framework for training. The MIND module can not only model the dynamics of the environment (e.g. 'what might happen if the agent passes through a door') but also help the agent to create a better understanding of the environment (e.g. 'The refrigerator is usually in the kitchen'). Such knowledge makes the agent a faster and better learner in locating a feasible policy with only a few trails. Furthermore, the MIND module can generate mental images that are treated as short-term subgoals by our proposed deep reinforcement framework. These mental images facilitate policy learning since short-term subgoals are easy to achieve and reusable. This yields better planning efficiency than other algorithms that learn a policy directly from primitive actions. Finally, the mental images visualize the agent's intentions in a way that human can understand, and this endows our agent's actions with more interpretability. The experimental results and further analysis prove that the agent with the MIND module is superior to its counterparts not only in EQA performance but in many other aspects such as route planning, behavioral interpretation, and the ability to generalize from a few examples.

Infrared images have a wide range of military and civilian applications, including night vision, surveillance, and robotics. However, high-resolution infrared detectors are difficult to fabricate and their manufacturing cost is expensive. In this paper, we present a cascaded architecture of deep neural networks with multiple receptive fields to increase the spatial resolution of infrared images by a large scale factor (x8). Instead of reconstructing a high-resolution image from its low-resolution version using a single complex deep network, the key idea of our approach is to set up a mid-point (scale x2) between scale x1 and x8 such that lost information can be divided into two components. Lost information within each component contains similar patterns thus can be more accurately recovered even using a simpler deep network. In our proposed cascaded architecture, two consecutive deep networks with different receptive fields are jointly trained through a multi-scale loss function. The first network with a large receptive field is applied to recover largescale structure information, while the second one uses a relatively smaller receptive field to reconstruct small-scale image details. Our proposed method is systematically evaluated using realistic infrared images. Compared with state-of-the-art super-resolution methods, our proposed cascaded approach achieves improved reconstruction accuracy using significantly fewer parameters.

In the development of a new product, the design team must describe the expected effects of the final products to potential users and stakeholders. However, existing prototyping tools can only present a product imperfectly, due to limitations at different levels. Specifically, the physical product model, which may be the product of 3D printing, could lack a visual interface; the presentation of the product through modeling software such as Rhinoceros 3D does not provide good realistic tactile perception; or the interface platforms, such as Axure RP, used to display the interactive effects differ from those to be used in the actual operation. Thus, we present the VPModel, a high-fidelity prototyping tool, able to integrate multiple prototyping methods simultaneously. It combines a touchable 3D-printed product model (3DPM) and a corresponding visualized virtual model, and the interactive interfaces are rendered synchronously in a mixed-reality device. Through the tangible, visual, and interactive demonstration, designers and normal users can each obtain a similar experience to the experience of the finished product. Furthermore, the VPModel also enhances design practices by enabling comparisons between modular models. However, the implementation of this system is a challenging task, which subsumes several fundamental problems as sub-tasks: object detection, real-time matching, hand-gesture detection and action recognition. To achieve the expected goals of the VPModel, this system uses physical hardware (a Microsoft MR HoloLens headset, a Leap Motion Controller, and a 3D printer) and existing machine learning algorithms. To evaluate our VPModel, we report the user experience of 16 participants, evaluated using a closed-ended questionnaire survey, a quantitative analysis of task performance, and a qualitative analysis of open-ended interviews. The results show a significant improvement in realism and enjoyment using the VPModel over the two traditional camera prototype approaches. In summary, the VPModel can be used to support design strategy and to convey design concepts fully and efficiently, which indicates a potential use for the VPModel in shortening product development cycles and reducing communication costs.

With the ever-increasing urbanization process, the traffic jam has become a common problem in the metropolises around the world, making the traffic speed prediction a crucial and fundamental task. This task is difficult due to the dynamic and intrinsic complexity of the traffic environment in urban cities, yet the emergence of crowd map query data sheds new light on it. In general, a burst of crowd map queries for the same destination in a short duration (called "hotspot'') could lead to traffic congestion. For example, queries of the Capital Gym burst on weekend evenings lead to traffic jams around the gym. However, unleashing the power of crowd map queries is challenging due to the innate spatiotemporal characteristics of the crowd queries. To bridge the gap, this paper firstly discovers hotspots underlying crowd map queries. These discovered hotspots address the spatiotemporal variations. Then Dest-ResNet (Deep spatiotemporal Residual Network) is proposed for hotspot traffic speed prediction. Dest-ResNet is a sequence learning framework that jointly deals with two sequences in different modalities, i.e., the traffic speed sequence and the query sequence. The main idea of Dest-ResNet is to learn to explain and amend the errors caused when the unimodal information is applied individually. In this way, Dest-ResNet addresses the temporal causal correlation between queries and the traffic speed. As a result, Dest-ResNet shows a 30% relative boost over the state-of-the-art methods on real-world datasets from Baidu Map.

Question answering is an important problem that aims to deliver specific answers to questions posed by humans in natural language. How to efficiently identify the exact answer with respect to a given question has become an active line of research. Previous approaches in factoid question answering tasks typically focus on modeling the semantic relevance or syntactic relationship between a given question and its corresponding answer. Most of these models suffer when a question contains very little content that is indicative of the answer. In this paper, we devise an architecture named the temporality-enhanced knowledge memory network (TE-KMN) and apply the model to a factoid question answering dataset from a trivia competition called quiz bowl. Unlike most of the existing approaches, our model encodes not only the content of questions and answers, but also the temporal cues in a sequence of ordered sentences which gradually remark the answer. Moreover, our model collaboratively uses external knowledge for a better understanding of a given question. The experimental results demonstrate that our method achieves better performance than several state-of-the-art methods.

In domain-specific NER, due to insufficient labeled training data, deep models usually fail to behave normally.In this paper, we proposed a novel Neural Inductive TEaching framework (NITE) to transfer knowledge from existing domain-specific NER models into an arbitrary deep neural network in a teacher-student training manner.NITE is a general framework that builds upon transfer learning and multiple instance learning, which collaboratively not only transfers knowledge to a deep student network but also reduces the noise from teachers.NITE can help deep learning methods to effectively utilize existing resources (i.e., models, labeled and unlabeled data) in a small domain.The experiment resulted on Disease NER proved that without using any labeled data, NITE can significantly boost the performance of a CNN-bidirectional LSTM-CRF NER neural network nearly over 30% in terms of F1-score.

Understanding the semantics in videos is a complex but crucial task in video analysis. This paper focuses on localizing category-independent events, actions or other semantics in an untrimmed video, referred as salient temporal proposal localization. Traditional methods like sliding window have a high computational cost due to the densely sampling of different video segments. We propose a reinforcement learning based method, which trains a localizer that learns a search policy that, instead of exploring every video segment, finds an optimal search path to locate a salient proposal based on the currently observing video segment in a tree structure, therefore reduces the number of video segments fed into the proposal detector. In each search step, a localizer is trained to iteratively select the next sub-region containing salient proposals to continue the search, and a proposal detector is trained to recognize salient proposal from the sub-regions. The experiments demonstrate that our method is able to precisely detect salient proposals with a comparable recall and with much fewer candidate windows.

Distant Supervision is a widely used approach for training relation extraction models. It generates noisy training samples by heuristically labeling a corpus using an existing knowledge base. Previous noise reduction methods for distant supervision fail to utilize information such as data credibility and sample confidence. In this paper, we proposed a novel neural framework, named ENCORE (External Neural COnstraints REgularized distant supervision), which allows an integration of other information for standard DS through regularizations under multiple external neural networks. In ENCORE, a teacher-student co-training mechanism is used to iterative distilling information from external neural networks to an existing relation extraction model. The experiment results demonstrated that without increasing any data or reshaping its original structure, ENCORE enhanced a CNN based relation extraction model for over 12%. The enhanced model also outperforms the state-of-the-art relation extraction method on the same dataset.

Named entity disambiguation (NED) is the task of linking mentions of ambiguous entities to their referenced entities in a knowledge base such as Wikipedia. We propose an approach to effectively disentangle the discriminative features in the manner of collaborative utilization of collective wisdom (via human-labeled crowd labels) and deep learning (via human-generated data) for the NED task. In particular, we devise a crowd model to elicit the underlying features (crowd features) from crowd labels that indicate a matching candidate for each mention, and then use the crowd features to fine-tune a dynamic convolutional neural network (DCNN). The learned DCNN is employed to obtain deep crowd features to enhance traditional hand-crafted features for the NED task. The proposed method substantially benefits from the utilization of crowd knowledge (via crowd labels) into a generic deep learning for the NED task. Experimental analysis demonstrates that the proposed approach is superior to the traditional hand-crafted features when enough crowd labels are gathered.

During the last decade, community-based question answering (CQA) sites have accumulated a vast amount of questions and their crowdsourced answers over time. How to efficiently identify the quality of answers that are relevant to a given question has become an active line of research in CQA. The major challenge of CQA is the accurate selection of high-quality answers w.r.t given questions. Previous approaches tend to model the semantic matching between individual pair of one question and its corresponding answer (how fitting an answer is to a posted question). However, these works ignore the temporal interactions between answers (how previous answers influence the late posted answers). For example, a rational user likely adapts others' opinions, revises his inclinations, and posts a more appropriate answer after understanding the given question and previously posted answers. As a result, this paper devises an architecture named Temporal Interaction and Causal Influence LSTM (TC-LSTM) to effectively leverage not only the causal influence between question-answer (how appropriate an answer is for a given question) but also the temporal interactions between answers-answer (how a high-quality answer gradually forms). In particular, long short-term memory (LSTM) is used to capture the explicit question-answer influence and the implicit answers-answer interactions. Experiments are conducted on SemEval 2015 CQA dataset for answer classification task and Baidu Zhidao Dataset for answer ranking task. The experimental results show the advantage of our model comparing with other state-of-the-art methods.

Search engine logs contain a large amount of users' click-through data that can be leveraged as implicit indicators of relevance. In this paper we address ad recommendation problem that finding and ranking the most relevant ads with respect to users' search queries. Due to the click sparsity, the conventional methods can hardly model the both inter- and intra-relations among users, queries and ads. We utilize the long-short term memory(LSTM) network to effectively encode two kinds of sequences: the (user, query) sequence and the query word sequence to represent users' query intention in a continuous vector space and decode them as distributions over ads respectively. Further more, we combine these two LSTM networks in an appropriate way to build up a more robust model referred as composite LSTM model(cLSTM) for ad recommendation. We evaluate the proposed cLSTM on real world click-through data set comparing with two baseline methods, the results demonstrate that our proposed model outperforms two baselines and mitigate the click sparsity problem to a certain degree.

In this paper, we propose an approach for generating rich fine-grained textual descriptions of images. In particular, we use an LSTM-in-LSTM (long short-term memory) architecture, which consists of an inner LSTM and an outer LSTM. The inner LSTM effectively encodes the long-range implicit contextual interaction between visual cues (i.e., the spatiallyconcurrent visual objects), while the outer LSTM generally captures the explicit multi-modal relationship between sentences and images (i.e., the correspondence of sentences and images). This architecture is capable of producing a long description by predicting one word at every time step conditioned on the previously generated word, a hidden vector (via the outer LSTM), and a context vector of fine-grained visual cues (via the inner LSTM). Our model outperforms state-of-theart methods on several benchmark datasets (Flickr8k, Flickr30k, MSCOCO) when used to generate long rich fine-grained descriptions of given images in terms of four different metrics (BLEU, CIDEr, ROUGE-L, and METEOR).

Summarization is desirable for efficient comprehension of an increasingly vast amount of data. A summary of multiple documents is a concise description of the main topic. Generally speaking, a topic delivers various aspects. For example, the natural disaster topic is likely to imply the aspects of casualties and rescue. Therefore, a good summary is expected to cover all the informative aspects of a topic in order to enhance both diversity and coverage of the topic. However, for the real-world data, the profile of aspects in a given topic (e.g., the number of the aspects as well as their appropriate describing sentences or images) is hardly specified in advance. To address this problem, this paper proposes an approach to learn the hidden aspects in the topics via a nonparametric Bayesian model for multimedia summarization, namely, aspect learning for multimedia summarization via nonparametric Bayesian (ALSNB). More specifically, we introduce the priors of beta-Bernoulli process and Dirichlet process into the traditional dictionary learning. As a result, the proposed approach is able to adaptively identify the particular aspects of an individual topic. The experimental results on several datasets for text summarization and image summarization show the superiority of the proposed ALSNB over other methods.

Cross-modal retrieval is a very hot research topic that is imperative to many applications involving multi-modal data. Discovering an appropriate representation for multi-modal data and learning a ranking function are essential to boost the cross-media retrieval. Motivated by the assumption that a compositional cross-modal semantic representation (pairs of images and text) is more attractive for cross-modal ranking, this paper exploits the existing image-text databases to optimize a ranking function for cross-modal retrieval, called deep compositional cross-modal learning to rank (C2MLR). In this paper, C2MLR considers learning a multi-modal embedding from the perspective of optimizing a pairwise ranking problem while enhancing both local alignment and global alignment. In particular, the local alignment (i.e., the alignment of visual objects and textual words) and the global alignment (i.e., the image-level and sentence-level alignment) are collaboratively utilized to learn the multi-modal embedding common space in a max-margin learning to rank manner. The experiments demonstrate the superiority of our proposed C2MLR due to its nature of multi-modal compositional embedding.

Cross-modal ranking is a research topic that is imperative to many applications involving multimodal data. Discovering a joint representation for multimodal data and learning a ranking function are essential in order to boost the cross-media retrieval (i.e., image-query-text or text-query-image). In this paper, we propose an approach to discover the latent joint representation of pairs of multimodal data (e.g., pairs of an image query and a text document) via a conditional random field and structural learning in a listwise ranking manner. We call this approach cross-modal learning to rank via latent joint representation (CML²R). In CML²R, the correlations between multimodal data are captured in terms of their sharing hidden variables (e.g., topics), and a hidden-topic-driven discriminative ranking function is learned in a listwise ranking manner. The experiments show that the proposed approach achieves a good performance in cross-media retrieval and meanwhile has the capability to learn the discriminative representation of multimodal data.

We propose selective supervised Latent Dirichlet Allocation (ssLDA) to boost the prediction performance of the widely studied supervised probabilistic topic models. We introduce a Bernoulli distribution for each word in one given document to selectthis word as a strongly or weakly discriminative one with respect to its assigned topic. The Bernoulli distribution is parameterized by the discrimination power of the word for its assigned topic. As a result, the document is represented as a “bag-of-selective-words” instead of the probabilistic “bag-of-topics” in the topic modeling domain or the flat “bag-of-words” in the traditional natural language processing domain to form a new perspective. Inheriting the general framework of supervised LDA (sLDA), ssLDA can also predict many types of response specified by a Gaussian Linear Model (GLM). Focusing on the utilization of this word selection mechanism for singe-label document classification in this paper, we conduct the variational inference for approximating the intractable posterior and derive a maximum-likelihood estimation of parameters in ssLDA. The experiments reported on textual documents show that ssLDA not only performs competitively over “state-of-the-art” classification approaches based on both the flat “bag-of-words” and probabilistic “bag-of-topics” representation in terms of classification performance, but also has the ability to discover the discrimination power of the words specified in the topics (compatible with our rational knowledge).

In this paper, structured visual feature learning aims at exploiting the intrinsic structural properties of mutually correlated multimedia collections (e.g., video frames or facial images) to learn a more effective feature representation for multimedia data classification. We pose structured visual feature learning as a problem of supervised tensor factorization (STF), which is capable of effectively learning multi-view visual features from structural tensorial multimedia data. In mathematics , STF is formulated as a joint optimization framework of probabilistic inference and$\epsilon $-insensitive support vector regression. As a result, the feature representation obtained by STF not only preserves the intrinsic multi-view structural information on tensorial multimedia data, but also includes the discriminative information derived from the max-margin learning process. Using the learned discriminative visual features, we conduct a set of multimedia classification experiments on several challenging datasets, including images and videos, which demonstrate the effectiveness of our method.

Nowadays, the prevalence of sharing tourist photos to online communities has created an increasing demand for mining discriminative architecture aspects from historic landmarks. Some previous researches have demonstrated that topic models could discover discriminative features represented by meaningful visual-topics. However, they seldom exploited the indicative function of geo-tags and the hierarchy in architecture characteristics. In order to utilize this information, we proposed a semi-supervised hierarchical topic modeling approach (namely, shTM). In our approach, every image could be represented by a probability distribution over selected geo-related visual-topics from a partly randomized topic tree. We evaluated our approach on a real-world dataset with over 26 thousand geo-informative photos from Flickr. Experiments show that shTM topics could reveal more discriminative aspects of a specific architecture than other well-known image features, such as HOG and SIFT, on the tasks of automatic photo categorization and geographical information retrieval.

Cross-media retrieval is a challenging problem in multimedia retrieval area. In the real-world, many applications involve multi-modal data, e.g., web pages containing both images and texts. How to utilize the intrinsic intra-modality and inter-modality similarity to learn the appropriate relationships of the data objects and provide efficient search across different modalities is the core of cross-media retrieval. Inspired by the fact that hashing methods well address the fast retrieval problem in the large-scale data settings, designing a cross-media hashing approach which can perform efficient retrieval over heterogenous high-dimensional feature spaces is highly desirable. In this paper, we propose a cross-media hashing approach based on kernel regression (abbreviated as KRCMH) to obtain the hash codes for the data objects across different modalities. The experiments on two real-world data sets show that KRCMH achieves superior cross-media retrieval performance comparing with the state-of-the-art methods.

The ever-increasing user-generated contents in social media and other web services make it highly desirable to discover opinions of users on all kinds of topics. Motivated by the assumption that individual word and paragraph in documents will deliver fine-grained (e.g., "laudatory", "annoyed" or "boring") and coarse-grained (e.g., positive, negative or neutral) sentiments about certain topics respectively, this paper focuses on a deeper thematic level to jointly disentangle fine-grained and coarse-grained opinions towards topics in terms of sentiment analysis, named as LDA with multi-grained sentiments (MgS-LDA). As a result, the proposed MgS-LDA not only discovers the topics in social media, but also identifies opinions about a given topic in terms of fine-grained and coarse-grained sentiment. Results of several experiments show that our proposed MgS-LDA achieves better performance on both sentimental classification and topic modeling than related methods.

Cross-media hashing, which conducts cross-media retrieval by embedding data from different modalities into a common low-dimensional hamming space, has attracted intensive attention in recent years. This is motivated by the facts a) the multi-modal data is widespread, e.g., the web images on Flickr are associated with tags, and b) hashing is an effective technique towards large-scale high-dimensional data processing, which is exactly the situation of cross-media retrieval. Inspired by recent advances in deep learning, we propose a cross-media hashing approach based on multi-modal neural networks. By restricting in the learning objective a) the hash codes for relevant cross-media data being similar, and b) the hash codes being discriminative for predicting the class labels, the learned Hamming space is expected to well capture the cross-media semantic relationships and to be semantically discriminative. The experiments on two real-world data sets show that our approach achieves superior cross-media retrieval performance compared with the state-of-the-art methods.

Hashing techniques have been extensively investigated to boost similarity search for large-scale high-dimensional data. Most of the existing approaches formulate the their objective as a pair-wise similarity-preserving problem. In this paper, we consider the hashing problem from the perspective of optimizing a list-wise learning to rank problem and propose an approach called List-Wise supervised Hashing (LWH). In LWH, the hash functions are optimized by employing structural SVM in order to explicitly minimize the ranking loss of the whole list-wise permutations instead of merely the point-wise or pair-wise supervision. We evaluate the performance of LWH on two real-world data sets. Experimental results demonstrate that our method obtains a significant improvement over the state-of-the-art hashing approaches due to both structural large margin and list-wise ranking pursuing in a supervised manner.

Learning hash functions across heterogenous high-dimensional features is very desirable for many applications involving multi-modal data objects. In this paper, we propose an approach to obtain the sparse codesets for the data objects across different modalities via joint multi-modal dictionary learning, which we call sparse multi-modal hashing (abbreviated as${\rm SM}^{2}{\rm H}$). In${\rm SM}^{2}{\rm H}$, both intra-modality similarity and inter-modality similarity are first modeled by a hypergraph, then multi-modal dictionaries are jointly learned by Hypergraph Laplacian sparse coding. Based on the learned dictionaries, the sparse codeset of each data object is acquired and conducted for multi-modal approximate nearest neighbor retrieval using a sensitive Jaccard metric. The experimental results show that${\rm SM}^{2}{\rm H}$outperforms other methods in terms of mAP and Percentage on two real-world data sets.

Non-negative tensor factorization (NTF) has attracted great attention in the machine learning community. In this paper, we extend traditional non-negative tensor factorization into a supervised discriminative decomposition, referred as Supervised Non-negative Tensor Factorization with Maximum-Margin Constraint(SNTFM2). SNTFM2 formulates the optimal discriminative factorization of non-negative tensorial data as a coupled least-squares optimization problem via a maximum-margin method. As a result, SNTFM2 not only faithfully approximates the tensorial data by additive combinations of the basis, but also obtains a strong generalization power to discriminative analysis (in particularfor classification in this paper). The experimental results show the superiority of our proposed model over state-of-the-art techniques on both toy and real world data sets.

Segments, such as sentence boundaries in texts or annotated regions in images, can be considered as useful structural constraints (i.e., priors) for unsupervised topic modeling. However, some segment units (e.g., words in texts or visual words in images) inside a given segment may be irrelevant to the topic of this segment due to their characteristics. This paper proposes a model called πLDA, which introduces a latent variable π into LDA, a traditional topic model, to capture the characteristic of each segment unit. That is to say, the πLDA model is conducted to determine whether a segment unit is assigned (or selected) to the topic embedded in its corresponding segment. Compared with other approaches that assume all the segment units in one segment to share a common topic, our proposed πLDA has the selective ability to discover the discriminative segment units (e.g., informative words or visual words). Experimental results and interpretations of them are presented for demonstrating the promising performance of our method.

Cross-modal retrieval is a classic research topic in multimedia information retrieval. The traditional approaches study the problem as a pairwise similarity function problem. In this paper, we consider this problem from a new perspective as a listwise ranking problem and propose a general cross-modal ranking algorithm to optimize the listwise ranking loss with a low rank embedding, which we call Latent Semantic Cross-Modal Ranking (LSCMR). The latent low-rank embedding space is discriminatively learned by structural large margin learning to optimize for certain ranking criteria directly. We evaluate LSCMR on the Wikipedia and NUS-WIDE dataset. Experimental results show that this method obtains significant improvements over the state-of-the-art methods.

Nowadays, vast amounts of multimedia data can be obtained across different collections (or domains). Therefore, it poses significant challenges for the utilization of those cross-collection data, for examples, the summarization of similarities and differences of data across different domains (e.g., CNN and NYT), as well as finding visually similar images across different visual domains (e.g., photos, paintings and hand-drawn sketches). In this paper, a supervised cross-collection Latent Dirichlet Allocation (scLDA) approach is proposed to utilize the data across different collections. As a natural extension of traditional Latent Dirichlet Allocation (LDA), scLDA not only takes the structural priors of different collections into consideration, but also exploits the category information. The strength of this work lies in integrating topic modeling, cross-domain learning and supervised learning together. We conduct scLDA for comparative text mining as well as classification of news articles and images from different collections. The results suggest that our proposed scLDA can generate meaningful collection-specific topics and achieves better retrieval accuracy than other related topic models.