High-quality long-context data is essential for training large language models (LLMs) capable of processing extensive documents, yet existing synthesis approaches using relevance-based aggregation face challenges of computational efficiency. We present LiteLong, a resource-efficient method for synthesizing long-context data through structured topic organization and multi-agent debate. Our approach leverages the BISAC book classification system to provide a comprehensive hierarchical topic organization, and then employs a debate mechanism with multiple LLMs to generate diverse, high-quality topics within this structure. For each topic, we use lightweight BM25 retrieval to obtain relevant documents and concatenate them into 128K-token training samples. Experiments on HELMET and Ruler benchmarks demonstrate that LiteLong achieves competitive long-context performance and can seamlessly integrate with other long-dependency enhancement methods. LiteLong makes high-quality long-context data synthesis more accessible by reducing both computational and data engineering costs, facilitating further research in long-context language training.

Vision Transformers (ViTs) have emerged as state-of-the-art models for various vision tasks recently. However, their heavy computation costs remain daunting for resource-limited devices. To address this, researchers have dedicated themselves to compressing redundant information in ViTs for acceleration. However, existing approaches generally sparsely drop redundant image tokens by token pruning or brutally remove channels by channel pruning, leading to a sub-optimal balance between model performance and inference speed. Moreover, they struggle when transferring compressed models to downstream vision tasks that require the spatial structure of images, such as semantic segmentation. To tackle these issues, we propose CAIT, a joint compression method for ViTs that achieves a harmonious blend of high accuracy, fast inference speed, and favorable transferability to downstream tasks. Specifically, we introduce an asymmetric token merging (ATME) strategy to effectively integrate neighboring tokens. It can successfully compress redundant token information while preserving the spatial structure of images. On top of it, we further design a consistent dynamic channel pruning (CDCP) strategy to dynamically prune unimportant channels in ViTs. Thanks to CDCP, insignificant channels in multi-head self-attention modules of ViTs can be pruned uniformly, significantly enhancing the model compression. Extensive experiments on multiple benchmark datasets show that our proposed method can achieve state-of-the-art performance across various ViTs.

Byte Pair Encoding (BPE) serves as a foundation method for text tokenization in the Natural Language Processing (NLP) field. Despite its wide adoption, the original BPE algorithm harbors an inherent flaw: it inadvertently introduces a frequency imbalance for tokens in the text corpus. Since BPE iteratively merges the most frequent token pair in the text corpus to generate a new token and keeps all generated tokens in the vocabulary, it unavoidably holds tokens that primarily act as components of a longer token and appear infrequently on their own. We term such tokens as Scaffold Tokens. Due to their infrequent occurrences in the text corpus, Scaffold Tokens pose a learning imbalance issue. To address that issue, we propose Scaffold-BPE, which incorporates a dynamic scaffold token removal mechanism by parameter-free, computation-light, and easy-to-implement modifications to the original BPE method. This novel approach ensures the exclusion of low-frequency Scaffold Tokens from the token representations for given texts, thereby mitigating the issue of frequency imbalance and facilitating model training. On extensive experiments across language modeling and even machine translation, Scaffold-BPE consistently outperforms the original BPE, well demonstrating its effectiveness.

Segment Anything Model (SAM) has made great progress in anomaly segmentation tasks due to its impressive generalization ability. However, existing methods that directly apply SAM through prompting often overlook the domain shift issue, where SAM performs well on natural images but struggles in industrial scenarios. Parameter-Efficient Fine-Tuning (PEFT) offers a promising solution, but it may yield suboptimal performance by not adequately addressing the perception challenges during adaptation to anomaly images. In this paper, we propose a novel Self-Perception Tuning (SPT) method, aiming to enhance SAM's perception capability for anomaly segmentation. The SPT method incorporates a self-drafting tuning strategy, which generates an initial coarse draft of the anomaly mask, followed by a refinement process. Additionally, a visual-relation-aware adapter is introduced to improve the perception of discriminative relational information for mask generation. Extensive experimental results on several benchmark datasets demonstrate that our SPT method can significantly outperform baseline methods, validating its effectiveness.

Processing long input remains a significant challenge for large language models (LLMs) due to the scarcity of large-scale long-context training data and the high computational cost of training models for extended context windows.In this paper, we propose Adaptive Grouped Positional Encoding (AdaGroPE), a training-free, plug-and-play method to enhance long-context understanding in existing LLMs.AdaGroPE progressively increases the reuse count of relative positions as the distance grows and dynamically adapts the positional encoding mapping to sequence length, thereby fully exploiting the range of pre-trained position embeddings.Its design is consistent with the principles of rotary position embedding (RoPE) and aligns with human perception of relative distance, enabling robust performance in realworld settings with variable-length inputs.Extensive experiments across various benchmarks demonstrate that our AdaGroPE consistently achieves state-of-the-art performance, surpassing baseline methods and even outperforming LLMs inherently designed for long-context processing on certain tasks.

Recently, Large language models (LLMs) have revolutionized Natural Language Processing (NLP). Pretrained LLMs, due to limited training context size, struggle with handling long token sequences, limiting their performance on various downstream tasks. Current solutions toward long context modeling often employ multi-stage continual pertaining, which progressively increases the effective context length through several continual pretraining stages. However, those approaches require extensive manual tuning and human expertise. In this paper, we introduce a novel single-stage continual pretraining method, Head-Adaptive Rotary Position Embedding (HARPE), to equip LLMs with long context modeling capabilities while simplifying the training process. Our HARPE leverages different Rotary Position Embedding (RoPE) base frequency values across different attention heads and directly trains LLMs on the target context length. Extensive experiments on 4 language modeling benchmarks, including the latest RULER benchmark, demonstrate that HARPE excels in understanding and integrating long-context tasks with single-stage training, matching and even outperforming existing multi-stage methods. Our results highlight that HARPE successfully breaks the stage barrier for training LLMs with long context modeling capabilities.

Vision network designs, including Convolutional Neural Networks and Vision Transformers, have significantly advanced the field of computer vision. Yet, their complex computations pose challenges for practical deployments, particularly in real-time applications. To tackle this issue, researchers have explored various lightweight and efficient network designs. However, existing lightweight models predominantly leverage self-attention mechanisms and convolutions for token mixing. This dependence brings limitations in effectiveness and efficiency in the perception and aggregation processes of lightweight networks, hindering the balance between performance and efficiency under limited computational budgets. In this paper, we draw inspiration from the dynamic heteroscale vision ability inherent in the efficient human vision system and propose a "See Large, Focus Small" strategy for lightweight vision network design. We introduce LS (Large-Small) convolution, which combines large-kernel perception and small-kernel aggregation. It can efficiently capture a wide range of perceptual information and achieve precise feature aggregation for dynamic and complex visual representations, thus enabling proficient processing of visual information. Based on LS convolution, we present LSNet, a new family of lightweight models. Extensive experiments demonstrate that LSNet achieves superior performance and efficiency over existing lightweight networks in various vision tasks. Codes and models are available at https://github.com/jameslahm/lsnet.

Minxuan Lv, Zhenpeng Su, Leiyu Pan, Yizhe Xiong, Zijia Lin, Hui Chen, Wei Zhou, Jungong Han, Guiguang Ding, Wenwu Ou, Di Zhang, Kun Gai, Songlin Hu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Yizhe Xiong, Xiansheng Chen, Xin Ye, Hui Chen, Zijia Lin, Haoran Lian, Zhenpeng Su, Wei Huang, Jianwei Niu, Jungong Han, Guiguang Ding. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The prevalent use of Byte Pair Encoding (BPE) in Large Language Models (LLMs) facilitates robust handling of subword units and avoids issues of out-of-vocabulary words. Despite its success, a critical challenge persists: long tokens, rich in semantic information, have fewer occurrences in tokenized datasets compared to short tokens, which can result in imbalanced learning issue across different tokens. To address that, we propose LBPE, which prioritizes long tokens during the encoding process. LBPE generates tokens according to their descending order of token length rather than their ranks in the vocabulary, granting longer tokens higher priority during the encoding process. Consequently, LBPE smooths the frequency differences between short and long tokens, and thus mitigates the learning imbalance. Extensive experiments across diverse language modeling tasks demonstrate that LBPE consistently outperforms the original BPE, well demonstrating its effectiveness.

Large language models (LLMs) with extended context windows have made significant strides yet remain a challenge due to the scarcity of long documents. Existing methods tend to synthesize long-context data but lack a clear mechanism to reinforce the long-range dependency modeling. To address this limitation, we propose NExtLong, a novel framework for synthesizing long-context data through Negative document Extension. NExtLong decomposes a document into multiple meta-chunks and extends the context by interleaving hard negative distractors retrieved from pretraining corpora. This approach compels the model to discriminate long-range dependent context from distracting content, enhancing its ability to model long-range dependencies. Extensive experiments demonstrate that NExtLong achieves significant performance improvements on the HELMET and RULER benchmarks compared to existing long-context synthesis approaches and leading models, which are trained on non-synthetic long documents. These findings highlight NExtLong's ability to reduce reliance on non-synthetic long documents, making it an effective framework for developing advanced long-context LLMs.

Vision-language models (VLMs) exhibit remarkable zero-shot capabilities but struggle with distribution shifts in downstream tasks when labeled data is unavailable, which has motivated the development of Test-Time Adaptation (TTA) to improve VLMs' performance during inference without annotations. Among various TTA approaches, cache-based methods show promise by preserving historical knowledge from low-entropy samples in a dynamic cache and fostering efficient adaptation. However, these methods face two critical reliability challenges: (1) entropy often becomes unreliable under distribution shifts, causing error accumulation in the cache and degradation in adaptation performance; (2) the final predictions may be unreliable due to inflexible decision boundaries that fail to accommodate large downstream shifts. To address these challenges, we propose a Reliable Test-time Adaptation (ReTA) method that integrates two complementary strategies to enhance reliability from two perspectives. First, to mitigate the unreliability of entropy as a sample selection criterion for cache construction, we introduce Consistency-aware Entropy Reweighting (CER), which incorporates consistency constraints to weight entropy during cache updating. While conventional approaches rely solely on low entropy for cache prioritization and risk introducing noise, our method leverages predictive consistency to maintain a high-quality cache and facilitate more robust adaptation. Second, we present Diversity-driven Distribution Calibration (DDC), which models class-wise text embeddings as multivariate Gaussian distributions, enabling adaptive decision boundaries for more accurate predictions across visually diverse content. Extensive experiments demonstrate that ReTA consistently outperforms state-of-the-art methods, particularly under real-world distribution shifts.

Zhenpeng Su, Xing W, Zijia Lin, Yizhe Xiong, Minxuan Lv, Guangyuan Ma, Hui Chen, Songlin Hu, Guiguang Ding. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Xinhao Xu, Hui Chen, Mengyao Lyu, Sicheng Zhao, Yizhe Xiong, Zijia Lin, Jungong Han, Guiguang Ding. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

High-quality long-context instruction data is essential for aligning long-context large language models (LLMs). Despite the public release of models like Qwen and Llama, their long-context instruction data remains proprietary. Human annotation is costly and challenging, while template-based synthesis methods limit scale, diversity, and quality. We introduce LongMagpie, a self-synthesis framework that automatically generates large-scale long-context instruction data. Our key insight is that aligned long-context LLMs, when presented with a document followed by special tokens preceding a user turn, auto-regressively generate contextually relevant queries. By harvesting these document-query pairs and the model's responses, LongMagpie produces high-quality instructions without human effort. Experiments on HELMET, RULER, and Longbench v2 demonstrate that LongMagpie achieves leading performance on long-context tasks while maintaining competitive performance on short-context tasks, establishing it as a simple and effective approach for open, diverse, and scalable long-context instruction data synthesis.

Recently, large vision-language models (LVLMs) have rapidly gained popularity for their strong generation and reasoning capabilities given diverse multimodal inputs. However, these models incur significant computational and memory overhead during inference, which greatly hinders the efficient deployment in practical scenarios. The extensive key-value (KV) cache, necessitated by the lengthy input and output sequences, notably contributes to the high inference cost. Based on this, recent works have investigated ways to reduce the KV cache size for higher efficiency. Although effective, they generally overlook the distinct importance distributions of KV vectors across layers and maintain the same cache size for each layer during the next token prediction. This results in the significant contextual information loss for certain layers, leading to notable performance decline. To address this, we present PrefixKV. It reframes the challenge of determining KV cache sizes for all layers into the task of searching for the optimal global prefix configuration. With an adaptive layer-wise KV retention recipe based on binary search, the maximum contextual information can thus be preserved in each layer, facilitating the generation. Extensive experiments demonstrate that our method achieves the state-of-the-art performance compared with others. It exhibits superior inference efficiency and generation quality trade-offs, showing promising potential for practical applications. Code is available at https://github.com/THU-MIG/PrefixKV.

Recently, video-language understanding has achieved great success through large-scale pre-training. However, data scarcity remains a prevailing challenge. This study quantitatively reveals an "impossible trinity" among data quantity, diversity, and quality in pre-training datasets. Recent efforts seek to refine large-scale, diverse ASR datasets compromised by low quality through synthetic annotations. These methods successfully refine the original annotations by leveraging useful information in multimodal video content (frames, tags, ASR transcripts, etc.). Nevertheless, they struggle to mitigate noise within synthetic annotations and lack scalability as the dataset size expands. To address these issues, we introduce the Video DataFlywheel framework, which iteratively refines video annotations with improved noise control methods. For iterative refinement, we first leverage a video-language model to generate synthetic annotations, resulting in a refined dataset. Then, we pre-train on it and fine-tune on human refinement examples for a stronger model. These processes are repeated for continuous improvement. For noise control, we present AdaTaiLr, a novel method that requires weaker assumptions on noise distribution. This method proves more effective in large datasets and offers theoretical guarantees. The combination of iterative refinement and AdaTaiLr can achieve better scalability in video-language understanding. Extensive experiments show that our framework outperforms existing data refinement baselines, delivering a 3% performance boost and improving dataset quality with minimal diversity loss. Furthermore, our refined dataset facilitates significant improvements in various video-language understanding tasks, including video question answering and text-video retrieval.

Prompt tuning achieves superior performance across a wide range of tasks, including multi-label zero-shot classification. Existing approaches employ multiple prompts to acquire comprehensive knowledge from categories, demonstrating state-of-the-art performance and significant computational efficiency. However, two main challenges still exist in these methods that impede the full potential of generalization. First, the class imbalance is not carefully addressed. Despite some efforts to adopt re-weighted loss functions to alleviate the positive-negative imbalance, such strategies tend to exacerbate the class imbalance by over-suppression of labels with fewer samples and overfitting to dominant classes. Second, the multi-prompt methods neglect the interactions between prompts during parameter optimization, underestimating the potential of prompts and leading to suboptimal performance. To address these issues, we present a novel framework named Dynamic Regulation in Prompt Tuning (DAR-Prompt). DAR-Prompt introduces three dynamic components: semantic regulator and debiased regulator to address the class imbalance, along with contrastive gradient regularization to enhance feature separation through prompt interactions during the backward pass. Specifically, the semantic regulator generates class-adaptive thresholds to compensate for tail classes and mitigate over-suppression, while the debiased regulator focuses on learning biased classes by rectifying overconfident predictions. Moreover, we apply dynamic regularization to the gradient update directions of prompts to promote orthogonality, thereby enhancing feature distinctiveness. Extensive experiments on several benchmarks show that our method can achieve state-of-the-art performance, well demonstrating its effectiveness and superiority. Code is available at https://github.com/Evelyn1ywliang/DAR-Prompt.

Large language models have demonstrated exceptional capability in natural language understanding and generation. However, their generation speed is limited by the inherently sequential nature of their decoding process, posing challenges for real-time applications. This paper introduces Lexical Unit Decoding (LUD), a novel decoding methodology implemented in a data-driven manner, accelerating the decoding process without sacrificing output quality. The core of our approach is the observation that a pre-trained language model can confidently predict multiple contiguous tokens, forming the basis for a lexical unit, in which these contiguous tokens could be decoded in parallel. Extensive experiments validate that our method substantially reduces decoding time while maintaining generation quality, i.e., 33% speed up on natural language generation with no quality loss, and 30% speed up on code generation with a negligible quality loss of 3%. Distinctively, LUD requires no auxiliary models and does not require changes to existing architectures. It can also be integrated with other decoding acceleration methods, thus achieving an even more pronounced inference efficiency boost. We posit that the foundational principles of LUD could define a new decoding paradigm for future language models, enhancing their applicability for a broader spectrum of applications. All codes are be publicly available at https://github.com/tjunlp-lab/Lexical-Unit-Decoding-LUD-.

Recently, lightweight Vision Transformers (ViTs) demon-strate superior performance and lower latency, compared with lightweight Convolutional Neural Networks (CNNs), on resource-constrained mobile devices. Researchers have discovered many structural connections be-tween lightweight ViTs and lightweight CNNs. However, the notable architectural disparities in the block structure, macro, and micro designs between them have not been adequately examined. In this study, we revisit the efficient design of lightweight CNNs from ViT perspective and emphasize their promising prospect for mobile devices. Specifically, we incrementally enhance the mobile-friendliness of a standard lightweight CNN, i.e., MobileNetV3, by integrating the efficient architectural designs of lightweight ViTs. This ends up with a new family of pure lightweight CNNs, namely RepViT. Extensive experiments show that RepViT outperforms existing state-of-the-art lightweight ViTs and exhibits favorable latency in various vision tasks. Notably, on ImageNet, RepViT achieves over 80% top-1 accuracy with 1.0 ms latency on an iPhone 12, which is the first time for a lightweight model, to the best of our knowledge. Besides, when RepViT meets SAM, our RepViT-SAM can achieve nearly 10x faster inference than the advanced MobileSAM. Codes and models are available at https://github.com/THU-MIG/RepViT.

Unsupervised anomaly detection (AD) aims to train robust detection models using only normal samples, while can generalize well to unseen anomalies. Recent research focuses on a unified unsupervised AD setting in which only one model is trained for all classes, i.e., n-class-one-model paradigm. Feature-reconstruction-based methods achieve state-of-the-art performance in this scenario. However, existing methods often suffer from a lack of sufficient contextual awareness, thereby compromising the quality of the reconstruction. To address this issue, we introduce a novel Reconstruction as Sequence (RAS) method, which enhances the contextual correspondence during feature reconstruction from a sequence modeling perspective. In particular, based on the transformer technique, we integrate a specialized RASFormer block into RAS. This block enables the capture of spatial relationships among different image regions and enhances sequential dependencies throughout the reconstruction process. By incorporating the RASFormer block, our RAS method achieves superior contextual awareness capabilities, leading to remarkable performance. Experimental results show that our RAS significantly outperforms competing methods, well demonstrating the effectiveness and superiority of our method. Our code is available at https://github.com/Nothingtolose9979/RAS

With the growing popularity of general-purpose Large Language Models (LLMs), comes a need for more global explanations of model behaviors.Concept-based explanations arise as a promising avenue for explaining high-level patterns learned by LLMs.Yet their evaluation poses unique challenges, especially due to their non-local nature and high dimensional representation in a model's hidden space.Current methods approach concepts from different perspectives, lacking a unified formalization.This makes evaluating the core measures of concepts, namely faithfulness or readability, challenging.To bridge the gap, we introduce a formal definition of concepts generalizing to diverse concept-based explanations' settings.Based on this, we quantify the faithfulness of a concept explanation via perturbation.We ensure adequate perturbation in the highdimensional space for different concepts via an optimization problem.Readability is approximated via an automatic and deterministic measure, quantifying the coherence of patterns that maximally activate a concept while aligning with human understanding.Finally, based on measurement theory, we apply a metaevaluation method for evaluating these measures, generalizable to other types of explanations or tasks as well.Extensive experimental analysis has been conducted to inform the selection of explanation evaluation measures.

Over the past years, YOLOs have emerged as the predominant paradigm in the field of real-time object detection owing to their effective balance between computational cost and detection performance. Researchers have explored the architectural designs, optimization objectives, data augmentation strategies, and others for YOLOs, achieving notable progress. However, the reliance on the non-maximum suppression (NMS) for post-processing hampers the end-to-end deployment of YOLOs and adversely impacts the inference latency. Besides, the design of various components in YOLOs lacks the comprehensive and thorough inspection, resulting in noticeable computational redundancy and limiting the model's capability. It renders the suboptimal efficiency, along with considerable potential for performance improvements. In this work, we aim to further advance the performance-efficiency boundary of YOLOs from both the post-processing and the model architecture. To this end, we first present the consistent dual assignments for NMS-free training of YOLOs, which brings the competitive performance and low inference latency simultaneously. Moreover, we introduce the holistic efficiency-accuracy driven model design strategy for YOLOs. We comprehensively optimize various components of YOLOs from both the efficiency and accuracy perspectives, which greatly reduces the computational overhead and enhances the capability. The outcome of our effort is a new generation of YOLO series for real-time end-to-end object detection, dubbed YOLOv10. Extensive experiments show that YOLOv10 achieves the state-of-the-art performance and efficiency across various model scales. For example, our YOLOv10-S is 1.8$\times$ faster than RT-DETR-R18 under the similar AP on COCO, meanwhile enjoying 2.8$\times$ smaller number of parameters and FLOPs. Compared with YOLOv9-C, YOLOv10-B has 46\% less latency and 25\% fewer parameters for the same performance. Code and models are available at https://github.com/THU-MIG/yolov10.

Masked auto-encoder pre-training has emerged as a prevalent technique for initializing and enhancing dense retrieval systems. It generally utilizes additional Transformer decoder blocks to provide sustainable supervision signals and compress contextual information into dense representations. However, the underlying reasons for the effectiveness of such a pre-training technique remain unclear. The usage of additional Transformer-based decoders also incurs significant computational costs. In this study, we aim to shed light on this issue by revealing that masked auto-encoder (MAE) pre-training with enhanced decoding significantly improves the term coverage of input tokens in dense representations, compared to vanilla BERT checkpoints. Building upon this observation, we propose a modification to the traditional MAE by replacing the decoder of a masked auto-encoder with a completely simplified Bag-of-Word prediction task. This modification enables the efficient compression of lexical signals into dense representations through unsupervised pre-training. Remarkably, our proposed method achieves state-of-the-art retrieval performance on several large-scale retrieval benchmarks without requiring any additional parameters, which provides a 67% training speed-up compared to standard masked auto-encoder pre-training with enhanced decoding.

Dense passage retrieval aims to retrieve the relevant passages of a query from a large corpus based on dense representations (i.e., vectors) of the query and the passages. Recent studies have explored improving pre-trained language models to boost dense retrieval performance. This paper proposes CoT-MAE (ConTextual Masked Auto-Encoder), a simple yet effective generative pre-training method for dense passage retrieval. CoT-MAE employs an asymmetric encoder-decoder architecture that learns to compress the sentence semantics into a dense vector through self-supervised and context-supervised masked auto-encoding. Precisely, self-supervised masked auto-encoding learns to model the semantics of the tokens inside a text span, and context-supervised masked auto-encoding learns to model the semantical correlation between the text spans. We conduct experiments on large-scale passage retrieval benchmarks and show considerable improvements over strong baselines, demonstrating the high efficiency of CoT-MAE. Our code is available at https://github.com/caskcsg/ir/tree/main/cotmae.

Recently, methods have been developed to improve the performance of dense passage retrieval by using context-supervised pre-training.These methods simply consider two passages from the same document to be relevant, without taking into account the potential negative impacts of weakly correlated pairs.Thus, this paper proposes query-as-context pre-training, a simple yet effective pre-training technique to alleviate the issue.Query-as-context pretraining assumes that the query derived from a passage is more likely to be relevant to that passage and forms a passage-query pair.These passage-query pairs are then used in contrastive or generative context-supervised pre-training.The pre-trained models are evaluated on largescale passage retrieval benchmarks and out-ofdomain zero-shot benchmarks.Experimental results show that query-as-context pre-training brings considerable gains for retrieval performances, demonstrating its effectiveness and efficiency.

Unsupervised domain adaptation aims to transfer knowledge from a fully-labeled source domain to an unlabeled target domain. However, in real-world scenarios, providing abundant labeled data even in the source domain can be infeasible due to the difficulty and high expense of annotation. To address this issue, recent works consider the Few-shot Unsupervised Domain Adaptation (FUDA) where only a few source samples are labeled, and conduct knowledge transfer via self-supervised learning methods. Yet existing methods generally overlook that the sparse label setting hinders learning reliable source knowledge for transfer. Additionally, the learning difficulty difference in target samples is different but ignored, leaving hard target samples poorly classified. To tackle both deficiencies, in this paper, we propose a novel Confidence-based Visual Dispersal Transfer learning method (C-VisDiT) for FUDA. Specifically, C-VisDiT consists of a cross-domain visual dispersal strategy that transfers only high-confidence source knowledge for model adaptation and an intra-domain visual dispersal strategy that guides the learning of hard target samples with easy ones. We conduct extensive experiments on Office-31, Office-Home, VisDA-C, and Domain- Net benchmark datasets and the results demonstrate that the proposed C-VisDiT significantly outperforms state-of- the-art FUDA methods. Our code is available at https://github.com/Bostoncake/C-VisDiT.

Collecting full annotations to construct multi-label datasets is difficult and labor-consuming. As an effective solution to relieve the annotation burden, single positive multi-label learning (SPML) draws increasing attention from both academia and industry. It only annotates each image with one positive label, leaving other labels unobserved. Therefore, existing methods strive to explore the cue of unobserved labels to compensate for the insufficiency of label supervision. Though achieving promising performance, they generally consider labels independently, leaving out the inherent hierarchical semantic relationship among labels which reveals that labels can be clustered into groups. In this paper, we propose a hierarchical prompt learning method with a novel Hierarchical Semantic Prompt Network (HSPNet) to harness such hierarchical semantic relationships using a large-scale pretrained vision and language model, i.e., CLIP, for SPML. We first introduce a Hierarchical Conditional Prompt (HCP) strategy to grasp the hierarchical label-group dependency. Then we equip a Hierarchical Graph Convolutional Network (HGCN) to capture the high-order inter-label and inter-group dependencies. Comprehensive experiments and analyses on several benchmark datasets show that our method significantly outperforms the state-of-the-art methods, well demonstrating its superiority and effectiveness. Our code will be available at https://github.com/jameslahm/HSPNet.

Despite the great success achieved by prevailing binary local descriptors, they are still suffering from two problems: 1) vulnerable to the geometric transformations; 2) lack of an effective treatment to the highly-correlated bits that are generated by directly applying the scheme of image hashing. To tackle both limitations, we propose an unsupervised Transformation-invariant Binary Local Descriptor learning method (TBLD). Specifically, the transformation invariance of binary local descriptors is ensured by projecting the original patches and their transformed counterparts into an identical high-dimensional feature space and an identical low-dimensional descriptor space simultaneously. Meanwhile, it enforces the dissimilar image patches to have distinctive binary local descriptors. Moreover, to reduce high correlations between bits, we propose a bottom-up learning strategy, termed Adversarial Constraint Module, where low-coupling binary codes are introduced externally to guide the learning of binary local descriptors. With the aid of the Wasserstein loss, the framework is optimized to encourage the distribution of the generated binary local descriptors to mimic that of the introduced low-coupling binary codes, eventually making the former more low-coupling. Experimental results on three benchmark datasets well demonstrate the superiority of the proposed method over the state-of-the-art methods. The project page is available at https://github.com/yoqim/TBLD.

While the performance of crowd counting via deep learning has been improved dramatically in the recent years, it remains an ingrained problem due to cluttered backgrounds and varying scales of people within an image. In this paper, we propose a Shallow feature based Dense Attention Network (SDANet) for crowd counting from still images, which diminishes the impact of backgrounds via involving a shallow feature based attention model, and meanwhile, captures multi-scale information via densely connecting hierarchical image features. Specifically, inspired by the observation that backgrounds and human crowds generally have noticeably different responses in shallow features, we decide to build our attention model upon shallow-feature maps, which results in accurate background-pixel detection. Moreover, considering that the most representative features of people across different scales can appear in different layers of a feature extraction network, to better keep them all, we propose to densely connect hierarchical image features of different layers and subsequently encode them for estimating crowd density. Experimental results on three benchmark datasets clearly demonstrate the superiority of SDANet when dealing with different scenarios. Particularly, on the challenging UCF_CC_50 dataset, our method outperforms other existing methods by a large margin, as is evident from a remarkable 11.9% Mean Absolute Error (MAE) drop of our SDANet.

For languages with no annotated resources, transferring knowledge from rich-resource languages is an effective solution for named entity recognition (NER). While all existing methods directly transfer from source-learned model to a target language, in this paper, we propose to fine-tune the learned model with a few similar examples given a test case, which could benefit the prediction by leveraging the structural and semantic information conveyed in such similar examples. To this end, we present a meta-learning algorithm to find a good model parameter initialization that could fast adapt to the given test case and propose to construct multiple pseudo-NER tasks for meta-training by computing sentence similarities. To further improve the model's generalization ability across different languages, we introduce a masking scheme and augment the loss function with an additional maximum term during meta-training. We conduct extensive experiments on cross-lingual named entity recognition with minimal resources over five target languages. The results show that our approach significantly outperforms existing state-of-the-art methods across the board.

To better tackle the named entity recognition (NER) problem on languages with little/no labeled data, cross-lingual NER must effectively leverage knowledge learned from source languages with rich labeled data.Previous works on cross-lingual NER are mostly based on label projection with pairwise texts or direct model transfer.However, such methods either are not applicable if the labeled data in the source languages is unavailable, or do not leverage information contained in unlabeled data in the target language.In this paper, we propose a teacher-student learning method to address such limitations, where NER models in the source languages are used as teachers to train a student model on unlabeled data in the target language.The proposed method works for both single-source and multi-source crosslingual NER.For the latter, we further propose a similarity measuring method to better weight the supervision from different teacher models.Extensive experiments for 3 target languages on benchmark datasets well demonstrate that our method outperforms existing state-of-theart methods for both single-source and multisource cross-lingual NER.

Enabling bi-directional retrieval of images and texts is important for understanding the correspondence between vision and language. Existing methods leverage the attention mechanism to explore such correspondence in a fine-grained manner. However, most of them consider all semantics equally and thus align them uniformly, regardless of their diverse complexities. In fact, semantics are diverse (i.e. involving different kinds of semantic concepts), and humans usually follow a latent structure to combine them into understandable languages. It may be difficult to optimally capture such sophisticated correspondences in existing methods. In this paper, to address such a deficiency, we propose an Iterative Matching with Recurrent Attention Memory (IMRAM) method, in which correspondences between images and texts are captured with multiple steps of alignments. Specifically, we introduce an iterative matching scheme to explore such fine-grained correspondence progressively. A memory distillation unit is used to refine alignment knowledge from early steps to later ones. Experiment results on three benchmark datasets, i.e. Flickr8K, Flickr30K, and MS COCO, show that our IMRAM achieves state-of-the-art performance, well demonstrating its effectiveness. Experiments on a practical business advertisement dataset, named KWAI-AD, further validates the applicability of our method in practical scenarios.

Prior work in cross-lingual named entity recognition (NER) with no/little labeled data falls into two primary categories: model transfer- and data transfer-based methods. In this paper, we find that both method types can complement each other, in the sense that, the former can exploit context information via language-independent features but sees no task-specific information in the target language; while the latter generally generates pseudo target-language training data via translation but its exploitation of context information is weakened by inaccurate translations. Moreover, prior work rarely leverages unlabeled data in the target language, which can be effortlessly collected and potentially contains valuable information for improved results. To handle both problems, we propose a novel approach termed UniTrans to Unify both model and data Transfer for cross-lingual NER, and furthermore, leverage the available information from unlabeled target-language data via enhanced knowledge distillation. We evaluate our proposed UniTrans over 4 target languages on benchmark datasets. Our experimental results show that it substantially outperforms the existing state-of-the-art methods.

Despite the thrilling success achieved by existing binary descriptors, most of them are still in the mire of three limitations: 1) vulnerable to the geometric transformations; 2) incapable of preserving the manifold structure when learning binary codes; 3) NO guarantee to find the true match if multiple candidates happen to have the same Hamming distance to a given query. All these together make the binary descriptor less effective, given large-scale visual recognition tasks. In this paper, we propose a novel learning-based feature descriptor, namely Unsupervised Deep Binary Descriptor (UDBD), which learns transformation invariant binary descriptors via projecting the original data and their transformed sets into a joint binary space. Moreover, we involve a ℓ2,1-norm loss term in the binary embedding process to gain simultaneously the robustness against data noises and less probability of mistakenly flipping bits of the binary descriptor, on top of it, a graph constraint is used to preserve the original manifold structure in the binary space. Furthermore, a weak bit mechanism is adopted to find the real match from candidates sharing the same minimum Hamming distance, thus enhancing matching performance. Extensive experimental results on public datasets show the superiority of UDBD in terms of matching and retrieval accuracy over state-of-the-arts.

Cross-modality human behavior analysis has attracted much attention from both academia and industry. In this article, we focus on the cross-modality image-text retrieval problem for human behavior analysis, which can learn a common latent space for cross-modality data and thus benefit the understanding of human behavior with data from different modalities. Existing state-of-the-art cross-modality image-text retrieval models tend to be fine-grained region-word matching approaches, where they begin with measuring similarities for each image region or text word followed by aggregating them to estimate the global image-text similarity. However, it is observed that such fine-grained approaches often encounter the similarity bias problem, because they only consider matched text words for an image region or matched image regions for a text word for similarity calculation, but they totally ignore unmatched words/regions, which might still be salient enough to affect the global image-text similarity. In this article, we propose an Adaptive Confidence Matching Network (ACMNet), which is also a fine-grained matching approach, to effectively deal with such a similarity bias. Apart from calculating the local similarity for each region(/word) with its matched words(/regions), ACMNet also introduces a confidence score for the local similarity by leveraging the global text(/image) information, which is expected to help measure the semantic relatedness of the region(/word) to the whole text(/image). Moreover, ACMNet also incorporates the confidence scores together with the local similarities in estimating the global image-text similarity. To verify the effectiveness of ACMNet, we conduct extensive experiments and make comparisons with state-of-the-art methods on two benchmark datasets, i.e., Flickr30k and MS COCO. Experimental results show that the proposed ACMNet can outperform the state-of-the-art methods by a clear margin, which well demonstrates the effectiveness of the proposed ACMNet in human behavior analysis and the reasonableness of tackling the mentioned similarity bias issue.

The dominant approaches for named entity recognitionm (NER) mostly adopt complex recurrent neural networks (RNN), e.g., long-short-term-memory (LSTM). However, RNNs are limited by their recurrent nature in terms of computational efficiency. In contrast, convolutional neural networks (CNN) can fully exploit the GPU parallelism with their feedforward architectures. However, little attention has been paid to performing NER with CNNs, mainly owing to their difficulties in capturing the long-term context information in a sequence. In this paper, we propose a simple but effective CNN-based network for NER, i.e., gated relation network (GRN), which is more capable than common CNNs in capturing long-term context. Specifically, in GRN we firstly employ CNNs to explore the local context features of each word. Then we model the relations between words and use them as gates to fuse local context features into global ones for predicting labels. Without using recurrent layers that process a sentence in a sequential manner, our GRN allows computations to be performed in parallel across the entire sentence. Experiments on two benchmark NER datasets (i.e., CoNLL2003 and Ontonotes 5.0) show that, our proposed GRN can achieve state-of-the-art performance with or without external knowledge. It also enjoys lower time costs to train and test.

Cross-modal image-text retrieval has been a long-standing challenge in the multimedia community. Existing methods explore various complicated embedding spaces to assess the semantic similarity between a given image-text pair, but consider no/little about the consistency across them. To remedy this situation, we introduce the idea of semantic consistency for learning various embedding spaces jointly. Specifically, similar to the previous works, we start by constructing two different embedding spaces, namely the image-grounded embedding space and the text-grounded embedding space. However, instead of learning these two embedding spaces separately, we incorporate a semantic consistency constraint in the common ranking objective function such that both embedding spaces can be learned simultaneously and benefit from each other to gain performance improvement. We conduct extensive experiments on three benchmark datasets, \ie Flickr8k, Flickr30k and MS COCO. Results show that our model outperforms the state-of-the-art models on all three datasets, which can well demonstrate the effectiveness and superiority of the introduction of semantic consistency. Our source code is released at: \urlhttps://github.com/HuiChen24/SemanticConsistency.

Chatbots have drawn significant attention of late in both industry and academia. For most task completion bots in the industry, human intervention is the only means of avoiding mistakes in complex real-world cases. However, to the best of our knowledge, there is no existing research work modeling the collaboration between task completion bots and human workers. In this paper, we introduce CoChat, a dialog management framework to enable effective collaboration between bots and human workers. In CoChat, human workers can introduce new actions at any time to handle previously unseen cases. We propose a memory-enhanced hierarchical RNN (MemHRNN) to handle the one-shot learning challenges caused by instantly introducing new actions in CoChat. Extensive experiments on real-world datasets well demonstrate that CoChat can relieve most of the human workers’ workload, and get better user satisfaction rates comparing to other state-of-the-art frameworks.

Despite the fact that attribute-based approaches and attention-based approaches have been proven to be effective in image captioning, most attribute-based approaches simply predict attributes independently without taking the co-occurrence dependencies among attributes into account. Besides, most attention-based captioning models directly leverage the feature map extracted from CNN, in which many features may be redundant in relation to the image content. In this paper, we focus on training a good attribute-inference model via the recurrent neural network (RNN) for image captioning, where the co-occurrence dependencies among attributes can be maintained. The uniqueness of our inference model lies in the usage of a RNN with the visual attention mechanism to \textit{observe} the image before generating captions. Additionally, it is noticed that compact and attribute-driven features will be more useful for the attention-based captioning model. To this end, we extract the context feature for each attribute, and guide the captioning model adaptively attend to these context features. We verify the effectiveness and superiority of the proposed approach over the other captioning approaches by conducting massive experiments and comparisons on MS COCO image captioning dataset.

Despite its great success, matrix factorization based cross-modality hashing suffers from two problems: 1) there is no engagement between feature learning and binarization; and 2) most existing methods impose the relaxation strategy by discarding the discrete constraints when learning the hash function, which usually yields suboptimal solutions. In this paper, we propose a novel multimodal hashing framework, referred as Unsupervised Deep Cross-Modal Hashing (UDCMH), for multimodal data search in a self-taught manner via integrating deep learning and matrix factorization with binary latent factor models. On one hand, our unsupervised deep learning framework enables the feature learning to be jointly optimized with the binarization. On the other hand, the hashing system based on the binary latent factor models can generate unified binary codes by solving a discrete-constrained objective function directly with no need for a relaxation step. Moreover, novel Laplacian constraints are incorporated into the objective function, which allow to preserve not only the nearest neighbors that are commonly considered in the literature but also the farthest neighbors of data, even if the semantic labels are not available. Extensive experiments on multiple datasets highlight the superiority of the proposed framework over several state-of-the-art baselines.

To make the problem of multilabel classification with many classes more tractable, in recent years, academia has seen efforts devoted to performing label space dimension reduction (LSDR). Specifically, LSDR encodes high-dimensional label vectors into low-dimensional code vectors lying in a latent space, so as to train predictive models at much lower costs. With respect to the prediction, it performs classification for any unseen instance by recovering a label vector from its predicted code vector via a decoding process. In this paper, we propose a novel method, namely End-to-End Feature-aware label space Encoding (E2FE), to perform LSDR. Instead of requiring an encoding function like most previous works, E2FE directly learns a code matrix formed by code vectors of the training instances in an end-to-end manner. Another distinct property of E2FE is its feature awareness attributable to the fact that the code matrix is learned by jointly maximizing the recoverability of the label space and the predictability of the latent space. Based on the learned code matrix, E2FE further trains predictive models to map instance features into code vectors, and also learns a linear decoding matrix for efficiently recovering the label vector of any unseen instance from its predicted code vector. Theoretical analyses show that both the code matrix and the linear decoding matrix in E2FE can be efficiently learned. Moreover, similar to previous works, E2FE can be specified to learn an encoding function. And it can also be extended with kernel tricks to handle nonlinear correlations between the feature space and the latent space. Comprehensive experiments conducted on diverse benchmark data sets with many classes show consistent performance gains of E2FE over the state-of-the-art methods.

For efficiently retrieving nearest neighbors from large-scale multiview data, recently hashing methods are widely investigated, which can substantially improve query speeds. In this paper, we propose an effective probability-based semantics-preserving hashing (SePH) method to tackle the problem of cross-view retrieval. Considering the semantic consistency between views, SePH generates one unified hash code for all observed views of any instance. For training, SePH first transforms the given semantic affinities of training data into a probability distribution, and aims to approximate it with another one in Hamming space, via minimizing their Kullback-Leibler divergence. Specifically, the latter probability distribution is derived from all pair-wise Hamming distances between to-be-learnt hash codes of the training data. Then with learnt hash codes, any kind of predictive models like linear ridge regression, logistic regression, or kernel logistic regression, can be learnt as hash functions in each view for projecting the corresponding view-specific features into hash codes. As for out-of-sample extension, given any unseen instance, the learnt hash functions in its observed views can predict view-specific hash codes. Then by deriving or estimating the corresponding output probabilities with respect to the predicted view-specific hash codes, a novel probabilistic approach is further proposed to utilize them for determining a unified hash code. To evaluate the proposed SePH, we conduct extensive experiments on diverse benchmark datasets, and the experimental results demonstrate that SePH is reasonable and effective.

With the dramatic development of the Internet, how to exploit large-scale retrieval techniques for multimodal web data has become one of the most popular but challenging problems in computer vision and multimedia. Recently, hashing methods are used for fast nearest neighbor search in large-scale data spaces, by embedding high-dimensional feature descriptors into a similarity preserving Hamming space with a low dimension. Inspired by this, in this paper, we introduce a novel supervised cross-modality hashing framework, which can generate unified binary codes for instances represented in different modalities. Particularly, in the learning phase, each bit of a code can be sequentially learned with a discrete optimization scheme that jointly minimizes its empirical loss based on a boosting strategy. In a bitwise manner, hash functions are then learned for each modality, mapping the corresponding representations into unified hash codes. We regard this approach as cross-modality sequential discrete hashing (CSDH), which can effectively reduce the quantization errors arisen in the oversimplified rounding-off step and thus lead to high-quality binary codes. In the test phase, a simple fusion scheme is utilized to generate a unified hash code for final retrieval by merging the predicted hashing results of an unseen instance from different modalities. The proposed CSDH has been systematically evaluated on three standard data sets: Wiki, MIRFlickr, and NUS-WIDE, and the results show that our method significantly outperforms the state-of-the-art multimodality hashing techniques.

With benefits of low storage costs and high query speeds, hashing methods are widely researched for efficiently retrieving large-scale data, which commonly contains multiple views, e.g. a news report with images, videos and texts. In this paper, we study the problem of cross-view retrieval and propose an effective Semantics-Preserving Hashing method, termed SePH. Given semantic affinities of training data as supervised information, SePH transforms them into a probability distribution and approximates it with to-be-learnt hash codes in Hamming space via minimizing the Kullback-Leibler divergence. Then kernel logistic regression with a sampling strategy is utilized to learn the nonlinear projections from features in each view to the learnt hash codes. And for any unseen instance, predicted hash codes and their corresponding output probabilities from observed views are utilized to determine its unified hash code, using a novel probabilistic approach. Extensive experiments conducted on three benchmark datasets well demonstrate the effectiveness and reasonableness of SePH.

To tackle a multi-label classification problem with many classes, recently label space dimension reduction (LSDR) is proposed. It encodes the original label space to a low-dimensional latent space and uses a decoding process for recovery. In this paper, we propose a novel method termed FaIE to perform LSDR via Feature-aware Implicit label space Encoding. Unlike most previous work, the proposed FaIE makes no assumptions about the encoding process and directly learns a code matrix, i.e. the encoding result of some implicit encoding function, and a linear decoding matrix. To learn both matrices, FaIE jointly maximizes the recoverability of the original label space from the latent space, and the predictability of the latent space from the feature space, thus making itself feature-aware. FaIE can also be specified to learn an explicit encoding function, and extended with kernel tricks to handle non-linear correlations between the feature space and the latent space. Extensive experiments conducted on benchmark datasets well demonstrate its effectiveness.

Though widely utilized for facilitating image management, user-provided image tags are usually incomplete and insufficient to describe the whole semantic content of corresponding images, resulting in performance degradations in tag-dependent applications and thus necessitating effective tag completion methods. In this paper, we propose a novel scheme denoted as LSR for automatic image tag completion via image-specific and tag-specific Linear Sparse Reconstructions. Given an incomplete initial tagging matrix with each row representing an image and each column representing a tag, LSR optimally reconstructs each image (i.e. row) and each tag (i.e. column) with remaining ones under constraints of sparsity, considering image-image similarity, image-tag association and tag-tag concurrence. Then both image-specific and tag-specific reconstruction values are normalized and merged for selecting missing related tags. Extensive experiments conducted on both benchmark dataset and web images well demonstrate the effectiveness of the proposed LSR.

Though the field of image auto-annotation has been extensively researched, most previous work concentrated on the single-source problem, assuming that both labelled and unseen to-be-annotated images are from a single source (e.g. an identical website), while in practice they are generally collected from multiple sources (e.g. different websites). In that case, treating each source independently may suffer from the insufficiency of labelled data for model training, while merging with labelled images from other sources can bring risky biases to the source-specific model. In this paper, we propose a multi-task learning model to alleviate the multi-source image auto-annotation problem, with each task defined as performing auto-annotation for the corresponding source. Specifically, the proposed model trains annotation models for all sources in parallel with the introduction of inter-source structure regularizers and parameter constraints for sharing information and enhancing the overall performance. Experiments conducted on three different-source benchmark datasets and their combinations yield inspiring results and demonstrate that the proposed model can well utilize the shared information and relieve the risky biases.

In this paper, we propose a novel image auto-annotation model using tag-related random search over range-constrained visual neighbors of the to-be-annotated image. The proposed model, termed as TagSearcher, observes that the annotating performances of many previous visual-neighbor-based models are generally sensitive to the quantity setting of visual neighbors, and the probabilities for visual neighbors to be selected is better to be tag-dependent, meaning that each candidate tag can have its own trustworthy part of visual neighbors for score prediction. And thus TagSearcher uses a constrained range rather than an identical and fixed number of visual neighbors for auto-annotation. By performing a novel tag-related random search process over the graphical model made up of range-constrained visual neighbors, TagSearcher can find the trustworthy part for each candidate tag, and further utilize both visual similarities and tag correlations for score prediction. With the range constraint for visual neighbors and the tag-related random search process, TagSearcher can not only achieve satisfactory annotating performances, but also reduce the performance sensitivity. Experiments conducted on benchmark Corel5k well demonstrate its rationality and effectiveness.

In this paper, a novel approach based on recommendation model is proposed for automatic image annotation. For any to-be-annotated image, we first select some related images with tags from training dataset according to their visual similarity. And then we estimate the initial ratings for tags of the training images based on tag ranking method and construct a rating matrix. We also construct a trust matrix based on visual similarity with a k-NN strategy. Then a recommendation model is built on both matrices to rank candidate tags for the target image. The proposed approach is evaluated using two benchmark image datasets, and experimental results have indicated its effectiveness.