Bin Xiao

Aiyuan Yang, Bin Xiao, Bingning Wang et al. · pku

Large language models (LLMs) have demonstrated remarkable performance on a variety of natural language tasks based on just a few examples of natural language instructions, reducing the need for extensive feature engineering. However, most powerful LLMs are closed-source or limited in their capability for languages other than English. In this technical report, we present Baichuan 2, a series of large-scale multilingual language models containing 7 billion and 13 billion parameters, trained from scratch, on 2.6 trillion tokens. Baichuan 2 matches or outperforms other open-source models of similar size on public benchmarks like MMLU, CMMLU, GSM8K, and HumanEval. Furthermore, Baichuan 2 excels in vertical domains such as medicine and law. We will release all pre-training model checkpoints to benefit the research community in better understanding the training dynamics of Baichuan 2.

Kan Wu, Jinnian Zhang, Houwen Peng et al.

Vision transformer (ViT) recently has drawn great attention in computer vision due to its remarkable model capability. However, most prevailing ViT models suffer from huge number of parameters, restricting their applicability on devices with limited resources. To alleviate this issue, we propose TinyViT, a new family of tiny and efficient small vision transformers pretrained on large-scale datasets with our proposed fast distillation framework. The central idea is to transfer knowledge from large pretrained models to small ones, while enabling small models to get the dividends of massive pretraining data. More specifically, we apply distillation during pretraining for knowledge transfer. The logits of large teacher models are sparsified and stored in disk in advance to save the memory cost and computation overheads. The tiny student transformers are automatically scaled down from a large pretrained model with computation and parameter constraints. Comprehensive experiments demonstrate the efficacy of TinyViT. It achieves a top-1 accuracy of 84.8% on ImageNet-1k with only 21M parameters, being comparable to Swin-B pretrained on ImageNet-21k while using 4.2 times fewer parameters. Moreover, increasing image resolutions, TinyViT can reach 86.5% accuracy, being slightly better than Swin-L while using only 11% parameters. Last but not the least, we demonstrate a good transfer ability of TinyViT on various downstream tasks. Code and models are available at https://github.com/microsoft/Cream/tree/main/TinyViT.

Jianwei Yang, Chunyuan Li, Pengchuan Zhang et al.

Visual recognition is recently learned via either supervised learning on human-annotated image-label data or language-image contrastive learning with webly-crawled image-text pairs. While supervised learning may result in a more discriminative representation, language-image pretraining shows unprecedented zero-shot recognition capability, largely due to the different properties of data sources and learning objectives. In this work, we introduce a new formulation by combining the two data sources into a common image-text-label space. In this space, we propose a new learning paradigm, called Unified Contrastive Learning (UniCL) with a single learning objective to seamlessly prompt the synergy of two data types. Extensive experiments show that our UniCL is an effective way of learning semantically rich yet discriminative representations, universally for image recognition in zero-shot, linear-probe, fully finetuning and transfer learning scenarios. Particularly, it attains gains up to 9.2% and 14.5% in average on zero-shot recognition benchmarks over the language-image contrastive learning and supervised learning methods, respectively. In linear probe setting, it also boosts the performance over the two methods by 7.3% and 3.4%, respectively. Our study also indicates that UniCL stand-alone is a good learner on pure image-label data, rivaling the supervised learning methods across three image classification datasets and two types of vision backbones, ResNet and Swin Transformer. Code is available at https://github.com/microsoft/UniCL.

Mingyu Ding, Bin Xiao, Noel Codella et al.

In this work, we introduce Dual Attention Vision Transformers (DaViT), a simple yet effective vision transformer architecture that is able to capture global context while maintaining computational efficiency. We propose approaching the problem from an orthogonal angle: exploiting self-attention mechanisms with both "spatial tokens" and "channel tokens". With spatial tokens, the spatial dimension defines the token scope, and the channel dimension defines the token feature dimension. With channel tokens, we have the inverse: the channel dimension defines the token scope, and the spatial dimension defines the token feature dimension. We further group tokens along the sequence direction for both spatial and channel tokens to maintain the linear complexity of the entire model. We show that these two self-attentions complement each other: (i) since each channel token contains an abstract representation of the entire image, the channel attention naturally captures global interactions and representations by taking all spatial positions into account when computing attention scores between channels; (ii) the spatial attention refines the local representations by performing fine-grained interactions across spatial locations, which in turn helps the global information modeling in channel attention. Extensive experiments show our DaViT achieves state-of-the-art performance on four different tasks with efficient computations. Without extra data, DaViT-Tiny, DaViT-Small, and DaViT-Base achieve 82.8%, 84.2%, and 84.6% top-1 accuracy on ImageNet-1K with 28.3M, 49.7M, and 87.9M parameters, respectively. When we further scale up DaViT with 1.5B weakly supervised image and text pairs, DaViT-Gaint reaches 90.4% top-1 accuracy on ImageNet-1K. Code is available at https://github.com/dingmyu/davit.

Zhecan Wang, Noel Codella, Yen-Chun Chen et al. · microsoft-research

Cross-modal encoders for vision-language (VL) tasks are often pretrained with carefully curated vision-language datasets. While these datasets reach an order of 10 million samples, the labor cost is prohibitive to scale further. Conversely, unimodal encoders are pretrained with simpler annotations that are less cost-prohibitive, achieving scales of hundreds of millions to billions. As a result, unimodal encoders have achieved state-of-art (SOTA) on many downstream tasks. However, challenges remain when applying to VL tasks. The pretraining data is not optimal for cross-modal architectures and requires heavy computational resources. In addition, unimodal architectures lack cross-modal interactions that have demonstrated significant benefits for VL tasks. Therefore, how to best leverage pretrained unimodal encoders for VL tasks is still an area of active research. In this work, we propose a method to leverage unimodal vision and text encoders for VL tasks that augment existing VL approaches while conserving computational complexity. Specifically, we propose Multimodal Adaptive Distillation (MAD), which adaptively distills useful knowledge from pretrained encoders to cross-modal VL encoders. Second, to better capture nuanced impacts on VL task performance, we introduce an evaluation protocol that includes Visual Commonsense Reasoning (VCR), Visual Entailment (SNLI-VE), and Visual Question Answering (VQA), across a variety of data constraints and conditions of domain shift. Experiments demonstrate that MAD leads to consistent gains in the low-shot, domain-shifted, and fully-supervised conditions on VCR, SNLI-VE, and VQA, achieving SOTA performance on VCR compared to other single models pretrained with image-text data. Finally, MAD outperforms concurrent works utilizing pretrained vision encoder from CLIP. Code will be made available.

Ziyi Yang, Yuwei Fang, Chenguang Zhu et al. · gatech, stanford

Human intelligence is multimodal; we integrate visual, linguistic, and acoustic signals to maintain a holistic worldview. Most current pretraining methods, however, are limited to one or two modalities. We present i-Code, a self-supervised pretraining framework where users may flexibly combine the modalities of vision, speech, and language into unified and general-purpose vector representations. In this framework, data from each modality are first given to pretrained single-modality encoders. The encoder outputs are then integrated with a multimodal fusion network, which uses novel attention mechanisms and other architectural innovations to effectively combine information from the different modalities. The entire system is pretrained end-to-end with new objectives including masked modality unit modeling and cross-modality contrastive learning. Unlike previous research using only video for pretraining, the i-Code framework can dynamically process single, dual, and triple-modality data during training and inference, flexibly projecting different combinations of modalities into a single representation space. Experimental results demonstrate how i-Code can outperform state-of-the-art techniques on five video understanding tasks and the GLUE NLP benchmark, improving by as much as 11% and demonstrating the power of integrative multimodal pretraining.

Kan Wu, Houwen Peng, Zhenghong Zhou et al.

In this paper, we propose a novel cross-modal distillation method, called TinyCLIP, for large-scale language-image pre-trained models. The method introduces two core techniques: affinity mimicking and weight inheritance. Affinity mimicking explores the interaction between modalities during distillation, enabling student models to mimic teachers' behavior of learning cross-modal feature alignment in a visual-linguistic affinity space. Weight inheritance transmits the pre-trained weights from the teacher models to their student counterparts to improve distillation efficiency. Moreover, we extend the method into a multi-stage progressive distillation to mitigate the loss of informative weights during extreme compression. Comprehensive experiments demonstrate the efficacy of TinyCLIP, showing that it can reduce the size of the pre-trained CLIP ViT-B/32 by 50%, while maintaining comparable zero-shot performance. While aiming for comparable performance, distillation with weight inheritance can speed up the training by 1.4 - 7.8 $\times$ compared to training from scratch. Moreover, our TinyCLIP ViT-8M/16, trained on YFCC-15M, achieves an impressive zero-shot top-1 accuracy of 41.1% on ImageNet, surpassing the original CLIP ViT-B/16 by 3.5% while utilizing only 8.9% parameters. Finally, we demonstrate the good transferability of TinyCLIP in various downstream tasks. Code and models will be open-sourced at https://aka.ms/tinyclip.

Haoxuan You, Luowei Zhou, Bin Xiao et al.

Large-scale multi-modal contrastive pre-training has demonstrated great utility to learn transferable features for a range of downstream tasks by mapping multiple modalities into a shared embedding space. Typically, this has employed separate encoders for each modality. However, recent work suggests that transformers can support learning across multiple modalities and allow knowledge sharing. Inspired by this, we investigate a variety of Modality-Shared Contrastive Language-Image Pre-training (MS-CLIP) frameworks. More specifically, we question how many parameters of a transformer model can be shared across modalities during contrastive pre-training, and rigorously examine architectural design choices that position the proportion of parameters shared along a spectrum. In studied conditions, we observe that a mostly unified encoder for vision and language signals outperforms all other variations that separate more parameters. Additionally, we find that light-weight modality-specific parallel modules further improve performance. Experimental results show that the proposed MS-CLIP approach outperforms vanilla CLIP by up to 13\% relative in zero-shot ImageNet classification (pre-trained on YFCC-100M), while simultaneously supporting a reduction of parameters. In addition, our approach outperforms vanilla CLIP by 1.6 points in linear probing on a collection of 24 downstream vision tasks. Furthermore, we discover that sharing parameters leads to semantic concepts from different modalities being encoded more closely in the embedding space, facilitating the transferring of common semantic structure (e.g., attention patterns) from language to vision. Code is available at \href{https://github.com/Hxyou/MSCLIP}{URL}.

Bin Xiao, Haiping Wu, Weijian Xu et al.

We introduce Florence-2, a novel vision foundation model with a unified, prompt-based representation for a variety of computer vision and vision-language tasks. While existing large vision models excel in transfer learning, they struggle to perform a diversity of tasks with simple instructions, a capability that implies handling the complexity of various spatial hierarchy and semantic granularity. Florence-2 was designed to take text-prompt as task instructions and generate desirable results in text forms, whether it be captioning, object detection, grounding or segmentation. This multi-task learning setup demands large-scale, high-quality annotated data. To this end, we co-developed FLD-5B that consists of 5.4 billion comprehensive visual annotations on 126 million images, using an iterative strategy of automated image annotation and model refinement. We adopted a sequence-to-sequence structure to train Florence-2 to perform versatile and comprehensive vision tasks. Extensive evaluations on numerous tasks demonstrated Florence-2 to be a strong vision foundation model contender with unprecedented zero-shot and fine-tuning capabilities.

Jinnian Zhang, Houwen Peng, Kan Wu et al.

Vision Transformer (ViT) models have recently drawn much attention in computer vision due to their high model capability. However, ViT models suffer from huge number of parameters, restricting their applicability on devices with limited memory. To alleviate this problem, we propose MiniViT, a new compression framework, which achieves parameter reduction in vision transformers while retaining the same performance. The central idea of MiniViT is to multiplex the weights of consecutive transformer blocks. More specifically, we make the weights shared across layers, while imposing a transformation on the weights to increase diversity. Weight distillation over self-attention is also applied to transfer knowledge from large-scale ViT models to weight-multiplexed compact models. Comprehensive experiments demonstrate the efficacy of MiniViT, showing that it can reduce the size of the pre-trained Swin-B transformer by 48\%, while achieving an increase of 1.0\% in Top-1 accuracy on ImageNet. Moreover, using a single-layer of parameters, MiniViT is able to compress DeiT-B by 9.7 times from 86M to 9M parameters, without seriously compromising the performance. Finally, we verify the transferability of MiniViT by reporting its performance on downstream benchmarks. Code and models are available at here.

Meituan LongCat Team, Bin Xiao, Chao Wang et al.

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

Jingcan Duan, Bin Xiao, Siwei Wang et al.

Recently, graph anomaly detection on attributed networks has attracted growing attention in data mining and machine learning communities. Apart from attribute anomalies, graph anomaly detection also aims at suspicious topological-abnormal nodes that exhibit collective anomalous behavior. Closely connected uncorrelated node groups form uncommonly dense substructures in the network. However, existing methods overlook that the topology anomaly detection performance can be improved by recognizing such a collective pattern. To this end, we propose a new graph anomaly detection framework on attributed networks via substructure awareness (ARISE for abbreviation). Unlike previous algorithms, we focus on the substructures in the graph to discern abnormalities. Specifically, we establish a region proposal module to discover high-density substructures in the network as suspicious regions. The average node-pair similarity can be regarded as the topology anomaly degree of nodes within substructures. Generally, the lower the similarity, the higher the probability that internal nodes are topology anomalies. To distill better embeddings of node attributes, we further introduce a graph contrastive learning scheme, which observes attribute anomalies in the meantime. In this way, ARISE can detect both topology and attribute anomalies. Ultimately, extensive experiments on benchmark datasets show that ARISE greatly improves detection performance (up to 7.30% AUC and 17.46% AUPRC gains) compared to state-of-the-art attributed networks anomaly detection (ANAD) algorithms.

Xinhang Wan, Bin Xiao, Xinwang Liu et al.

Multi-view clustering (MVC) has gained broad attention owing to its capacity to exploit consistent and complementary information across views. This paper focuses on a challenging issue in MVC called the incomplete continual data problem (ICDP). In specific, most existing algorithms assume that views are available in advance and overlook the scenarios where data observations of views are accumulated over time. Due to privacy considerations or memory limitations, previous views cannot be stored in these situations. Some works are proposed to handle it, but all fail to address incomplete views. Such an incomplete continual data problem (ICDP) in MVC is tough to solve since incomplete information with continual data increases the difficulty of extracting consistent and complementary knowledge among views. We propose Fast Continual Multi-View Clustering with Incomplete Views (FCMVC-IV) to address it. Specifically, it maintains a consensus coefficient matrix and updates knowledge with the incoming incomplete view rather than storing and recomputing all the data matrices. Considering that the views are incomplete, the newly collected view might contain samples that have yet to appear; two indicator matrices and a rotation matrix are developed to match matrices with different dimensions. Besides, we design a three-step iterative algorithm to solve the resultant problem in linear complexity with proven convergence. Comprehensive experiments on various datasets show the superiority of FCMVC-IV.

Lingyuan Meng, Ke Liang, Bin Xiao et al.

Few-shot relation reasoning on knowledge graphs (FS-KGR) aims to infer long-tail data-poor relations, which has drawn increasing attention these years due to its practicalities. The pre-training of previous methods needs to manually construct the meta-relation set, leading to numerous labor costs. Self-supervised learning (SSL) is treated as a solution to tackle the issue, but still at an early stage for FS-KGR task. Moreover, most of the existing methods ignore leveraging the beneficial information from aliasing relations (AR), i.e., data-rich relations with similar contextual semantics to the target data-poor relation. Therefore, we proposed a novel Self-Supervised Learning model by leveraging Aliasing Relations to assist FS-KGR, termed SARF. Concretely, four main components are designed in our model, i.e., SSL reasoning module, AR-assisted mechanism, fusion module, and scoring function. We first generate the representation of the co-occurrence patterns in a generative manner. Meanwhile, the representations of aliasing relations are learned to enhance reasoning in the AR-assist mechanism. Besides, multiple strategies, i.e., simple summation and learnable fusion, are offered for representation fusion. Finally, the generated representation is used for scoring. Extensive experiments on three few-shot benchmarks demonstrate that SARF achieves state-of-the-art performance compared with other methods in most cases.

Zhenrong Shen, Xi Ouyang, Bin Xiao et al.

Lung nodule detection in chest X-ray (CXR) images is common to early screening of lung cancers. Deep-learning-based Computer-Assisted Diagnosis (CAD) systems can support radiologists for nodule screening in CXR. However, it requires large-scale and diverse medical data with high-quality annotations to train such robust and accurate CADs. To alleviate the limited availability of such datasets, lung nodule synthesis methods are proposed for the sake of data augmentation. Nevertheless, previous methods lack the ability to generate nodules that are realistic with the size attribute desired by the detector. To address this issue, we introduce a novel lung nodule synthesis framework in this paper, which decomposes nodule attributes into three main aspects including shape, size, and texture, respectively. A GAN-based Shape Generator firstly models nodule shapes by generating diverse shape masks. The following Size Modulation then enables quantitative control on the diameters of the generated nodule shapes in pixel-level granularity. A coarse-to-fine gated convolutional Texture Generator finally synthesizes visually plausible nodule textures conditioned on the modulated shape masks. Moreover, we propose to synthesize nodule CXR images by controlling the disentangled nodule attributes for data augmentation, in order to better compensate for the nodules that are easily missed in the detection task. Our experiments demonstrate the enhanced image quality, diversity, and controllability of the proposed lung nodule synthesis framework. We also validate the effectiveness of our data augmentation on greatly improving nodule detection performance.

Wenxuan Tu, Bin Xiao, Xinwang Liu et al.

With the development of various applications, such as social networks and knowledge graphs, graph data has been ubiquitous in the real world. Unfortunately, graphs usually suffer from being absent due to privacy-protecting policies or copyright restrictions during data collection. The absence of graph data can be roughly categorized into attribute-incomplete and attribute-missing circumstances. Specifically, attribute-incomplete indicates that a part of the attribute vectors of all nodes are incomplete, while attribute-missing indicates that the whole attribute vectors of partial nodes are missing. Although many efforts have been devoted, none of them is custom-designed for a common situation where both types of graph data absence exist simultaneously. To fill this gap, we develop a novel network termed Revisiting Initializing Then Refining (RITR), where we complete both attribute-incomplete and attribute-missing samples under the guidance of a novel initializing-then-refining imputation criterion. Specifically, to complete attribute-incomplete samples, we first initialize the incomplete attributes using Gaussian noise before network learning, and then introduce a structure-attribute consistency constraint to refine incomplete values by approximating a structure-attribute correlation matrix to a high-order structural matrix. To complete attribute-missing samples, we first adopt structure embeddings of attribute-missing samples as the embedding initialization, and then refine these initial values by adaptively aggregating the reliable information of attribute-incomplete samples according to a dynamic affinity structure. To the best of our knowledge, this newly designed method is the first unsupervised framework dedicated to handling hybrid-absent graphs. Extensive experiments on four datasets have verified that our methods consistently outperform existing state-of-the-art competitors.

Xiuli Bi, Bo Liu, Fan Yang et al.

As deep learning technology continues to evolve, the images yielded by generative models are becoming more and more realistic, triggering people to question the authenticity of images. Existing generated image detection methods detect visual artifacts in generated images or learn discriminative features from both real and generated images by massive training. This learning paradigm will result in efficiency and generalization issues, making detection methods always lag behind generation methods. This paper approaches the generated image detection problem from a new perspective: Start from real images. By finding the commonality of real images and mapping them to a dense subspace in feature space, the goal is that generated images, regardless of their generative model, are then projected outside the subspace. As a result, images from different generative models can be detected, solving some long-existing problems in the field. Experimental results show that although our method was trained only by real images and uses 99.9\% less training data than other deep learning-based methods, it can compete with state-of-the-art methods and shows excellent performance in detecting emerging generative models with high inference efficiency. Moreover, the proposed method shows robustness against various post-processing. These advantages allow the method to be used in real-world scenarios.

Bin Xiao, Lujun Gui, Lei Su et al.

Large Language Models (LLMs) frequently suffer from inefficiencies, largely attributable to the discord between the requirements of auto-regressive decoding and the architecture of contemporary GPUs. Recently, regressive lightweight speculative decoding has garnered attention for its notable efficiency improvements in text generation tasks. This approach utilizes a lightweight regressive draft model, like a Recurrent Neural Network (RNN) or a single transformer decoder layer, leveraging sequential information to iteratively predict potential tokens. Specifically, RNN draft models are computationally economical but tend to deliver lower accuracy, while attention decoder layer models exhibit the opposite traits. This paper presents Clover-2, an advanced iteration of Clover, an RNN-based draft model designed to achieve comparable accuracy to that of attention decoder layer models while maintaining minimal computational overhead. Clover-2 enhances the model architecture and incorporates knowledge distillation to increase Clover's accuracy and improve overall efficiency. We conducted experiments using the open-source Vicuna 7B and LLaMA3-Instruct 8B models. The results demonstrate that Clover-2 surpasses existing methods across various model architectures, showcasing its efficacy and robustness.

Bin Xiao, Murat Simsek, Burak Kantarci et al.

Tabular data in digital documents is widely used to express compact and important information for readers. However, it is challenging to parse tables from unstructured digital documents, such as PDFs and images, into machine-readable format because of the complexity of table structures and the missing of meta-information. Table Structure Recognition (TSR) problem aims to recognize the structure of a table and transform the unstructured tables into a structured and machine-readable format so that the tabular data can be further analysed by the down-stream tasks, such as semantic modeling and information retrieval. In this study, we hypothesize that a complicated table structure can be represented by a graph whose vertices and edges represent the cells and association between cells, respectively. Then we define the table structure recognition problem as a cell association classification problem and propose a conditional attention network (CATT-Net). The experimental results demonstrate the superiority of our proposed method over the state-of-the-art methods on various datasets. Besides, we investigate whether the alignment of a cell bounding box or a text-focused approach has more impact on the model performance. Due to the lack of public dataset annotations based on these two approaches, we further annotate the ICDAR2013 dataset providing both types of bounding boxes, which can be a new benchmark dataset for evaluating the methods in this field. Experimental results show that the alignment of a cell bounding box can help improve the Micro-averaged F1 score from 0.915 to 0.963, and the Macro-average F1 score from 0.787 to 0.923.

Marah Abdin, Jyoti Aneja, Hany Awadalla et al. · microsoft-research, stanford

We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. Our training dataset is a scaled-up version of the one used for phi-2, composed of heavily filtered publicly available web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide parameter-scaling results with a 7B, 14B models trained for 4.8T tokens, called phi-3-small, phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75%, 78% on MMLU, and 8.7, 8.9 on MT-bench). To enhance multilingual, multimodal, and long-context capabilities, we introduce three models in the phi-3.5 series: phi-3.5-mini, phi-3.5-MoE, and phi-3.5-Vision. The phi-3.5-MoE, a 16 x 3.8B MoE model with 6.6 billion active parameters, achieves superior performance in language reasoning, math, and code tasks compared to other open-source models of similar scale, such as Llama 3.1 and the Mixtral series, and on par with Gemini-1.5-Flash and GPT-4o-mini. Meanwhile, phi-3.5-Vision, a 4.2 billion parameter model derived from phi-3.5-mini, excels in reasoning tasks and is adept at handling both single-image and text prompts, as well as multi-image and text prompts.

Bin Xiao, Yakup Akkaya, Murat Simsek et al.

The global Information and Communications Technology (ICT) supply chain is a complex network consisting of all types of participants. It is often formulated as a Social Network to discuss the supply chain network's relations, properties, and development in supply chain management. Information sharing plays a crucial role in improving the efficiency of the supply chain, and datasheets are the most common data format to describe e-component commodities in the ICT supply chain because of human readability. However, with the surging number of electronic documents, it has been far beyond the capacity of human readers, and it is also challenging to process tabular data automatically because of the complex table structures and heterogeneous layouts. Table Structure Recognition (TSR) aims to represent tables with complex structures in a machine-interpretable format so that the tabular data can be processed automatically. In this paper, we formulate TSR as an object detection problem and propose to generate an intuitive representation of a complex table structure to enable structuring of the tabular data related to the commodities. To cope with border-less and small layouts, we propose a cost-sensitive loss function by considering the detection difficulty of each class. Besides, we propose a novel anchor generation method using the character of tables that columns in a table should share an identical height, and rows in a table should share the same width. We implement our proposed method based on Faster-RCNN and achieve 94.79% on mean Average Precision (AP), and consistently improve more than 1.5% AP for different benchmark models.

Bin Xiao, Murat Simsek, Burak Kantarci et al.

Due to the characteristics of Information and Communications Technology (ICT) products, the critical information of ICT devices is often summarized in big tabular data shared across supply chains. Therefore, it is critical to automatically interpret tabular structures with the surging amount of electronic assets. To transform the tabular data in electronic documents into a machine-interpretable format and provide layout and semantic information for information extraction and interpretation, we define a Table Structure Recognition (TSR) task and a Table Cell Type Classification (CTC) task. We use a graph to represent complex table structures for the TSR task. Meanwhile, table cells are categorized into three groups based on their functional roles for the CTC task, namely Header, Attribute, and Data. Subsequently, we propose a multi-task model to solve the defined two tasks simultaneously by using the text modal and image modal features. Our experimental results show that our proposed method can outperform state-of-the-art methods on ICDAR2013 and UNLV datasets.

Xuelong Dai, Kaisheng Liang, Bin Xiao

Unrestricted adversarial attacks present a serious threat to deep learning models and adversarial defense techniques. They pose severe security problems for deep learning applications because they can effectively bypass defense mechanisms. However, previous attack methods often directly inject Projected Gradient Descent (PGD) gradients into the sampling of generative models, which are not theoretically provable and thus generate unrealistic examples by incorporating adversarial objectives, especially for GAN-based methods on large-scale datasets like ImageNet. In this paper, we propose a new method, called AdvDiff, to generate unrestricted adversarial examples with diffusion models. We design two novel adversarial guidance techniques to conduct adversarial sampling in the reverse generation process of diffusion models. These two techniques are effective and stable in generating high-quality, realistic adversarial examples by integrating gradients of the target classifier interpretably. Experimental results on MNIST and ImageNet datasets demonstrate that AdvDiff is effective in generating unrestricted adversarial examples, which outperforms state-of-the-art unrestricted adversarial attack methods in terms of attack performance and generation quality.

Yanjie Li, Bin Xie, Songtao Guo et al.

Benefiting from the rapid development of deep learning, 2D and 3D computer vision applications are deployed in many safe-critical systems, such as autopilot and identity authentication. However, deep learning models are not trustworthy enough because of their limited robustness against adversarial attacks. The physically realizable adversarial attacks further pose fatal threats to the application and human safety. Lots of papers have emerged to investigate the robustness and safety of deep learning models against adversarial attacks. To lead to trustworthy AI, we first construct a general threat model from different perspectives and then comprehensively review the latest progress of both 2D and 3D adversarial attacks. We extend the concept of adversarial examples beyond imperceptive perturbations and collate over 170 papers to give an overview of deep learning model robustness against various adversarial attacks. To the best of our knowledge, we are the first to systematically investigate adversarial attacks for 3D models, a flourishing field applied to many real-world applications. In addition, we examine physical adversarial attacks that lead to safety violations. Last but not least, we summarize present popular topics, give insights on challenges, and shed light on future research on trustworthy AI.

Yanjie Li, Yiquan Li, Xuelong Dai et al.

2D face recognition has been proven insecure for physical adversarial attacks. However, few studies have investigated the possibility of attacking real-world 3D face recognition systems. 3D-printed attacks recently proposed cannot generate adversarial points in the air. In this paper, we attack 3D face recognition systems through elaborate optical noises. We took structured light 3D scanners as our attack target. End-to-end attack algorithms are designed to generate adversarial illumination for 3D faces through the inherent or an additional projector to produce adversarial points at arbitrary positions. Nevertheless, face reflectance is a complex procedure because the skin is translucent. To involve this projection-and-capture procedure in optimization loops, we model it by Lambertian rendering model and use SfSNet to estimate the albedo. Moreover, to improve the resistance to distance and angle changes while maintaining the perturbation unnoticeable, a 3D transform invariant loss and two kinds of sensitivity maps are introduced. Experiments are conducted in both simulated and physical worlds. We successfully attacked point-cloud-based and depth-image-based 3D face recognition algorithms while needing fewer perturbations than previous state-of-the-art physical-world 3D adversarial attacks.

Xuhao Ren, Mingyang Zhao, Ruichen Zhang et al.

With the rapid development of mobile computing technology, massive amounts of spatial data are continuously generated from various mobile terminals and sensing devices, such as smartphones, connected vehicles, and drones. Performing efficient distributed statistical analysis on this data is crucial for real-time mobile computing applications. However, the constrained and dynamic nature of mobile environments exacerbates the privacy challenge: centralizing sensitive data for analysis risks severe privacy leaks, while existing privacy-preserving techniques often introduce excessive overhead or inaccuracies In this paper, we design, implement, and evaluate the first system that supports efficient and privacy-preserving distribution statistics analysis for mobile spatial data. First, we propose eSpat-B, which leverages two non-colluding servers and a newly designed improved distributed point functions (DPF) with octree partitioning. Furthermore, considering the frequent updates of spatial data, we propose another more efficient scheme, eSpat+. The core idea of this scheme is to utilize a K-Dimensional tree for spatial partitioning, combine it with incremental DPF for performing statistics analysis, and design an efficient update algorithm. Security analysis demonstrates that our schemes effectively protect data privacy throughout the statistical process. Theoretical analysis and experimental results on real-world mobile trajectory datasets demonstrate that our proposed schemes achieve a reduction of approximately 1.2* in computation overhead, 20* in communication overhead, and maintain 100% accuracy.

Shuo Wang, Keke Gai, Jing Yu et al.

Vertical federated learning has garnered significant attention as it allows clients to train machine learning models collaboratively without sharing local data, which protects the client's local private data. However, existing VFL methods face challenges when dealing with heterogeneous local models among participants, which affects optimization convergence and generalization. To address this challenge, this paper proposes a novel approach called Vertical federated learning for training multiple Heterogeneous models (VFedMH). VFedMH focuses on aggregating the local embeddings of each participant's knowledge during forward propagation. To protect the participants' local embedding values, we propose an embedding protection method based on lightweight blinding factors. In particular, participants obtain local embedding using local heterogeneous models. Then the passive party, who owns only features of the sample, injects the blinding factor into the local embedding and sends it to the active party. The active party aggregates local embeddings to obtain global knowledge embeddings and sends them to passive parties. The passive parties then utilize the global embeddings to propagate forward on their local heterogeneous networks. However, the passive party does not own the sample labels, so the local model gradient cannot be calculated locally. To overcome this limitation, the active party assists the passive party in computing its local heterogeneous model gradients. Then, each participant trains their local model using the heterogeneous model gradients. The objective is to minimize the loss value of their respective local heterogeneous models. Extensive experiments are conducted to demonstrate that VFedMH can simultaneously train multiple heterogeneous models with heterogeneous optimization and outperform some recent methods in model performance.

Kaisheng Liang, Bin Xiao

Adversarial attacks can mislead deep neural networks (DNNs) by adding imperceptible perturbations to benign examples. The attack transferability enables adversarial examples to attack black-box DNNs with unknown architectures or parameters, which poses threats to many real-world applications. We find that existing transferable attacks do not distinguish between style and content features during optimization, limiting their attack transferability. To improve attack transferability, we propose a novel attack method called style-less perturbation (StyLess). Specifically, instead of using a vanilla network as the surrogate model, we advocate using stylized networks, which encode different style features by perturbing an adaptive instance normalization. Our method can prevent adversarial examples from using non-robust style features and help generate transferable perturbations. Comprehensive experiments show that our method can significantly improve the transferability of adversarial examples. Furthermore, our approach is generic and can outperform state-of-the-art transferable attacks when combined with other attack techniques.

Haifeng Zhang, Qinghui He, Xiuli Bi et al.

In recent years, the rapid development of generative artificial intelligence technology has significantly lowered the barrier to creating high-quality fake images, posing a serious challenge to information authenticity and credibility. Existing generated image detection methods typically enhance generalization through model architecture or network design. However, their generalization performance remains susceptible to data bias, as the training data may drive models to fit specific generative patterns and content rather than the common features shared by images from different generative models (asymmetric bias learning). To address this issue, we propose a Multi-dimensional Adversarial Feature Learning (MAFL) framework. The framework adopts a pretrained multimodal image encoder as the feature extraction backbone, constructs a real-fake feature learning network, and designs an adversarial bias-learning branch equipped with a multi-dimensional adversarial loss, forming an adversarial training mechanism between authenticity-discriminative feature learning and bias feature learning. By suppressing generation-pattern and content biases, MAFL guides the model to focus on the generative features shared across different generative models, thereby effectively capturing the fundamental differences between real and generated images, enhancing cross-model generalization, and substantially reducing the reliance on large-scale training data. Through extensive experimental validation, our method outperforms existing state-of-the-art approaches by 10.89% in accuracy and 8.57% in Average Precision (AP). Notably, even when trained with only 320 images, it can still achieve over 80% detection accuracy on public datasets.

Meituan LongCat Team, Bairui Wang, Bayan et al.

We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.

Bin Xiao, Chien-Liang Liu, Wen-Hoar Hsaio

Few-shot learning (FSL) has attracted considerable attention recently. Among existing approaches, the metric-based method aims to train an embedding network that can make similar samples close while dissimilar samples as far as possible and achieves promising results. FSL is characterized by using only a few images to train a model that can generalize to novel classes in image classification problems, but this setting makes it difficult to learn the visual features that can identify the images' appearance variations. The model training is likely to move in the wrong direction, as the images in an identical semantic class may have dissimilar appearances, whereas the images in different semantic classes may share a similar appearance. We argue that FSL can benefit from additional semantic features to learn discriminative feature representations. Thus, this study proposes a multi-task learning approach to view semantic features of label text as an auxiliary task to help boost the performance of the FSL task. Our proposed model uses word-embedding representations as semantic features to help train the embedding network and a semantic cross-attention module to bridge the semantic features into the typical visual modal. The proposed approach is simple, but produces excellent results. We apply our proposed approach to two previous metric-based FSL methods, all of which can substantially improve performance. The source code for our model is accessible from github.

Ping Chen, Xingpeng Zhang, Ye Li et al.

Naked eye recognition of age is usually based on comparison with the age of others. However, this idea is ignored by computer tasks because it is difficult to obtain representative contrast images of each age. Inspired by the transfer learning, we designed the Delta Age AdaIN (DAA) operation to obtain the feature difference with each age, which obtains the style map of each age through the learned values representing the mean and standard deviation. We let the input of transfer learning as the binary code of age natural number to obtain continuous age feature information. The learned two groups of values in Binary code mapping are corresponding to the mean and standard deviation of the comparison ages. In summary, our method consists of four parts: FaceEncoder, DAA operation, Binary code mapping, and AgeDecoder modules. After getting the delta age via AgeDecoder, we take the average value of all comparison ages and delta ages as the predicted age. Compared with state-of-the-art methods, our method achieves better performance with fewer parameters on multiple facial age datasets.

ZeBin Ji, Yang Hu, Xiuli Bi et al.

It is essential for understanding neural network decisions to interpret the functionality (also known as concepts) of neurons. Existing approaches describe neuron concepts by generating natural language descriptions, thereby advancing the understanding of the neural network's decision-making mechanism. However, these approaches assume that each neuron has well-defined functions and provides discriminative features for neural network decision-making. In fact, some neurons may be redundant or may offer misleading concepts. Thus, the descriptions for such neurons may cause misinterpretations of the factors driving the neural network's decisions. To address the issue, we introduce a verification of neuron functions, which checks whether the generated concept highly activates the corresponding neuron. Furthermore, we propose a Select-Hypothesize-Verify framework for interpreting neuron functionality. This framework consists of: 1) selecting activation samples that best capture a neuron's well-defined functional behavior through activation-distribution analysis; 2) forming hypotheses about concepts for the selected neurons; and 3) verifying whether the generated concepts accurately reflect the functionality of the neuron. Extensive experiments show that our method produces more accurate neuron concepts. Our generated concepts activate the corresponding neurons with a probability approximately 1.5 times that of the current state-of-the-art method.

Lujun Gui, Bin Xiao, Lei Su et al.

Lossless speculative decoding accelerates target large language model (LLM) inference by employing a lightweight draft model for generating tree-structured candidates, which are subsequently verified in parallel by the target LLM. Currently, effective approaches leverage feature-level rather than token-level autoregression within the draft model to facilitate more straightforward predictions and enhanced knowledge distillation. In this paper, we reassess these approaches and propose FSPAD (Feature Sampling and Partial Alignment Distillation for Lossless Speculative Decoding), which introduces two straightforward and effective components within the existing framework to boost lossless speculative decoding. Firstly, FSPAD utilizes token embeddings to sample features of the target LLM in high-dimensional space before feeding them into the draft model, due to the inherent uncertainty of the features preventing the draft model from obtaining the specific token output by the target LLM. Secondly, FSPAD introduces partial alignment distillation to weaken the draft model's connection between features and logits, aiming to reduce the conflict between feature alignment and logit confidence during training. Our experiments include both greedy and non-greedy decoding on the largest and smallest models from the Vicuna and LLaMA3-Instruct series, as well as tasks in multi-turn conversation, translation, summarization, question answering, mathematical reasoning, and retrieval-augmented generation. The results show that FSPAD outperforms the state-of-the-art method across all the aforementioned tasks and target LLMs.

Xiuli Bi, Die Xiao, Junchao Fan et al.

In recent years, Contrastive Language-Image Pretraining (CLIP) has been widely applied to Weakly Supervised Semantic Segmentation (WSSS) tasks due to its powerful cross-modal semantic understanding capabilities. This paper proposes a novel Semantic and Spatial Rectification (SSR) method to address the limitations of existing CLIP-based weakly supervised semantic segmentation approaches: over-activation in non-target foreground regions and background areas. Specifically, at the semantic level, the Cross-Modal Prototype Alignment (CMPA) establishes a contrastive learning mechanism to enforce feature space alignment across modalities, reducing inter-class overlap while enhancing semantic correlations, to rectify over-activation in non-target foreground regions effectively; at the spatial level, the Superpixel-Guided Correction (SGC) leverages superpixel-based spatial priors to precisely filter out interference from non-target regions during affinity propagation, significantly rectifying background over-activation. Extensive experiments on the PASCAL VOC and MS COCO datasets demonstrate that our method outperforms all single-stage approaches, as well as more complex multi-stage approaches, achieving mIoU scores of 79.5% and 50.6%, respectively.

Fengli Ran, Xiao Pu, Bo Liu et al.

Dataset distillation compresses large datasets into compact synthetic ones to reduce storage and computational costs. Among various approaches, distribution matching (DM)-based methods have attracted attention for their high efficiency. However, they often overlook the evolution of feature representations during training, which limits the expressiveness of synthetic data and weakens downstream performance. To address this issue, we propose Trajectory Guided Dataset Distillation (TGDD), which reformulates distribution matching as a dynamic alignment process along the model's training trajectory. At each training stage, TGDD captures evolving semantics by aligning the feature distribution between the synthetic and original dataset. Meanwhile, it introduces a distribution constraint regularization to reduce class overlap. This design helps synthetic data preserve both semantic diversity and representativeness, improving performance in downstream tasks. Without additional optimization overhead, TGDD achieves a favorable balance between performance and efficiency. Experiments on ten datasets demonstrate that TGDD achieves state-of-the-art performance, notably a 5.0% accuracy gain on high-resolution benchmarks.

Xu Ma, Xiyang Dai, Jianwei Yang et al.

In this work, we present efficient modulation, a novel design for efficient vision networks. We revisit the modulation mechanism, which operates input through convolutional context modeling and feature projection layers, and fuses features via element-wise multiplication and an MLP block. We demonstrate that the modulation mechanism is particularly well suited for efficient networks and further tailor the modulation design by proposing the efficient modulation (EfficientMod) block, which is considered the essential building block for our networks. Benefiting from the prominent representational ability of modulation mechanism and the proposed efficient design, our network can accomplish better trade-offs between accuracy and efficiency and set new state-of-the-art performance in the zoo of efficient networks. When integrating EfficientMod with the vanilla self-attention block, we obtain the hybrid architecture which further improves the performance without loss of efficiency. We carry out comprehensive experiments to verify EfficientMod's performance. With fewer parameters, our EfficientMod-s performs 0.6 top-1 accuracy better than EfficientFormerV2-s2 and is 25% faster on GPU, and 2.9 better than MobileViTv2-1.0 at the same GPU latency. Additionally, our method presents a notable improvement in downstream tasks, outperforming EfficientFormerV2-s by 3.6 mIoU on the ADE20K benchmark. Code and checkpoints are available at https://github.com/ma-xu/EfficientMod.

Jiuhai Chen, Jianwei Yang, Haiping Wu et al.

We present Florence-VL, a new family of multimodal large language models (MLLMs) with enriched visual representations produced by Florence-2, a generative vision foundation model. Unlike the widely used CLIP-style vision transformer trained by contrastive learning, Florence-2 can capture different levels and aspects of visual features, which are more versatile to be adapted to diverse downstream tasks. We propose a novel feature-fusion architecture and an innovative training recipe that effectively integrates Florence-2's visual features into pretrained LLMs, such as Phi 3.5 and LLama 3. In particular, we propose "depth-breath fusion (DBFusion)" to fuse the visual features extracted from different depths and under multiple prompts. Our model training is composed of end-to-end pretraining of the whole model followed by finetuning of the projection layer and the LLM, on a carefully designed recipe of diverse open-source datasets that include high-quality image captions and instruction-tuning pairs. Our quantitative analysis and visualization of Florence-VL's visual features show its advantages over popular vision encoders on vision-language alignment, where the enriched depth and breath play important roles. Florence-VL achieves significant improvements over existing state-of-the-art MLLMs across various multi-modal and vision-centric benchmarks covering general VQA, perception, hallucination, OCR, Chart, knowledge-intensive understanding, etc. To facilitate future research, our models and the complete training recipe are open-sourced. https://github.com/JiuhaiChen/Florence-VL

Yanjie Li, Mingxing Duan, Xuelong Dai et al.

Adversarial patch attacks can fool the face recognition (FR) models via small patches. However, previous adversarial patch attacks often result in unnatural patterns that are easily noticeable. Generating transferable and stealthy adversarial patches that can efficiently deceive the black-box FR models while having good camouflage is challenging because of the huge stylistic difference between the source and target images. To generate transferable, natural-looking, and stealthy adversarial patches, we propose an innovative two-stage attack called Adv-Inpainting, which extracts style features and identity features from the attacker and target faces, respectively and then fills the patches with misleading and inconspicuous content guided by attention maps. In the first stage, we extract multi-scale style embeddings by a pyramid-like network and identity embeddings by a pretrained FR model and propose a novel Attention-guided Adaptive Instance Normalization layer (AAIN) to merge them via background-patch cross-attention maps. The proposed layer can adaptively fuse identity and style embeddings by fully exploiting priority contextual information. In the second stage, we design an Adversarial Patch Refinement Network (APR-Net) with a novel boundary variance loss, a spatial discounted reconstruction loss, and a perceptual loss to boost the stealthiness further. Experiments demonstrate that our attack can generate adversarial patches with improved visual quality, better stealthiness, and stronger transferability than state-of-the-art adversarial patch attacks and semantic attacks.

Jinwu Hu, Yufeng Wang, Shuhai Zhang et al.

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

Hongyuan Yu, Cheng Wan, Xiyang Dai et al.

Real-time image segmentation demands architectures that preserve fine spatial detail while capturing global context under tight latency and memory budgets. Image segmentation is one of the most fundamental problems in computer vision and has drawn a lot of attention due to its vast applications in image understanding and autonomous driving. However, designing effective and efficient segmentation neural architectures is a labor-intensive process that may require numerous trials by human experts. In this paper, we address the challenge of integrating multi-head self-attention into high-resolution representation CNNs efficiently by leveraging architecture search. Manually replacing convolution layers with multi-head self-attention is non-trivial due to the costly overhead in memory to maintain high resolution. By contrast, we develop a multi-target multi-branch supernet method, which not only fully utilizes the advantages of high-resolution features but also finds the proper location for placing the multi-head self-attention module. Our search algorithm is optimized towards multiple objectives (e.g., latency and mIoU) and is capable of finding architectures on the approximate Pareto front with an arbitrary number of branches in a single search. We further present a series of models via the Hybrid Convolutional-Transformer Architecture Search (HyCTAS) method that searches for the best hybrid combination of lightweight convolution layers and memory-efficient self-attention layers between branches from different resolutions and fuses them at high resolution for both efficiency and effectiveness. On Cityscapes, ADE20K, and COCO, HyCTAS discovers competitive real-time models without ImageNet pretraining, delivering strong accuracy and latency trade-offs. Code and models are available at https://github.com/MarvinYu1995/HyCTAS.

Jinwu Hu, Wei Zhang, Yufeng Wang et al.

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

Yanjie Li, Kaisheng Liang, Bin Xiao

In recent research, adversarial attacks on person detectors using patches or static 3D model-based texture modifications have struggled with low success rates due to the flexible nature of human movement. Modeling the 3D deformations caused by various actions has been a major challenge. Fortunately, advancements in Neural Radiance Fields (NeRF) for dynamic human modeling offer new possibilities. In this paper, we introduce UV-Attack, a groundbreaking approach that achieves high success rates even with extensive and unseen human actions. We address the challenge above by leveraging dynamic-NeRF-based UV mapping. UV-Attack can generate human images across diverse actions and viewpoints, and even create novel actions by sampling from the SMPL parameter space. While dynamic NeRF models are capable of modeling human bodies, modifying clothing textures is challenging because they are embedded in neural network parameters. To tackle this, UV-Attack generates UV maps instead of RGB images and modifies the texture stacks. This approach enables real-time texture edits and makes the attack more practical. We also propose a novel Expectation over Pose Transformation loss (EoPT) to improve the evasion success rate on unseen poses and views. Our experiments show that UV-Attack achieves a 92.7% attack success rate against the FastRCNN model across varied poses in dynamic video settings, significantly outperforming the state-of-the-art AdvCamou attack, which only had a 28.5% ASR. Moreover, we achieve 49.5% ASR on the latest YOLOv8 detector in black-box settings. This work highlights the potential of dynamic NeRF-based UV mapping for creating more effective adversarial attacks on person detectors, addressing key challenges in modeling human movement and texture modification. The code is available at https://github.com/PolyLiYJ/UV-Attack.

Yan Zhang, Pengcheng Zheng, Chengxiao Zeng et al.

Remote Sensing (RS) image deblurring and Super-Resolution (SR) are common tasks in computer vision that aim at restoring RS image detail and spatial scale, respectively. However, real-world RS images often suffer from a complex combination of global low-resolution (LR) degeneration and local blurring degeneration. Although carefully designed deblurring and SR models perform well on these two tasks individually, a unified model that performs jointly RS image deblurring and super-resolution (JRSIDSR) task is still challenging due to the vital dilemma of reconstructing the global and local degeneration simultaneously. Additionally, existing methods struggle to capture the interrelationship between deblurring and SR processes, leading to suboptimal results. To tackle these issues, we give a unified theoretical analysis of RS images' spatial and blur degeneration processes and propose a dual-branch parallel network named AKMD-Net for the JRSIDSR task. AKMD-Net consists of two main branches: deblurring and super-resolution branches. In the deblurring branch, we design a pixel-adjustable kernel block (PAKB) to estimate the local and spatial-varying blur kernels. In the SR branch, a multi-domain attention block (MDAB) is proposed to capture the global contextual information enhanced with high-frequency details. Furthermore, we develop an adaptive feature fusion (AFF) module to model the contextual relationships between the deblurring and SR branches. Finally, we design an adaptive Wiener loss (AW Loss) to depress the prior noise in the reconstructed images. Extensive experiments demonstrate that the proposed AKMD-Net achieves state-of-the-art (SOTA) quantitative and qualitative performance on commonly used RS image datasets. The source code is publicly available at https://github.com/zpc456/AKMD-Net.

Xuelong Dai, Bin Xiao

Recent studies that incorporate geometric features and transformers into 3D point cloud feature learning have significantly improved the performance of 3D deep-learning models. However, their robustness against adversarial attacks has not been thoroughly explored. Existing attack methods primarily focus on white-box scenarios and struggle to transfer to recently proposed 3D deep-learning models. Even worse, these attacks introduce perturbations to 3D coordinates, generating unrealistic adversarial examples and resulting in poor performance against 3D adversarial defenses. In this paper, we generate high-quality adversarial point clouds using diffusion models. By using partial points as prior knowledge, we generate realistic adversarial examples through shape completion with adversarial guidance. The proposed adversarial shape completion allows for a more reliable generation of adversarial point clouds. To enhance attack transferability, we delve into the characteristics of 3D point clouds and employ model uncertainty for better inference of model classification through random down-sampling of point clouds. We adopt ensemble adversarial guidance for improved transferability across different network architectures. To maintain the generation quality, we limit our adversarial guidance solely to the critical points of the point clouds by calculating saliency scores. Extensive experiments demonstrate that our proposed attacks outperform state-of-the-art adversarial attack methods against both black-box models and defenses. Our black-box attack establishes a new baseline for evaluating the robustness of various 3D point cloud classification models.

Bin Xiao, Lei Su

In the realm of Large Language Model (LLM) inference, the inherent structure of transformer models coupled with the multi-GPU tensor parallelism strategy leads to a sequential execution of computation and communication. This results in substantial underutilization of computing resources during the communication phase. To mitigate this inefficiency, various techniques have been developed to optimize the use of computational power throughout the communication process. These strategies primarily involve overlapping matrix computations and communications, as well as interleaving micro-batches across different requests. Nonetheless, these approaches either fall short of achieving ideal overlap or impose certain limitations on their application. To overcome these challenges, this paper introduces a novel strategy for computation-communication overlap that operates at the sequence level. This method not only enhances the degree of overlap but also minimizes the constraints on its applicability. Experimental evaluations conducted using 30b/70b models have demonstrated significant improvements in efficiency. Specifically, the proposed technique has been shown to reduce time consumption by approximately 35% on 4090 GPU and by roughly 15% on A800 GPU during the prefill stage of LLM inference.

ZhiChao Yan, Hui Xue, Yi Zhu et al.

Accurate segmentation of lesions in pancreatic endoscopic ultrasound (EUS) images is crucial for effective diagnosis and treatment. However, the collection of enough crisp EUS images for effective diagnosis is arduous. Recently, domain adaptation (DA) has been employed to address these challenges by leveraging related knowledge from other domains. Most DA methods only focus on multi-view representations of the same organ, which makes it still tough to clearly depict the tumor lesion area with limited semantic information. Although transferring homogeneous similarity from different organs could benefit the issue, there is a lack of relevant work due to the enormous domain gap between them. To address these challenges, we propose the Cross-Organ Tumor Segmentation Networks (COTS-Nets), consisting of a universal network and an auxiliary network. The universal network utilizes boundary loss to learn common boundary information of different tumors, enabling accurate delineation of tumors in EUS despite limited and low-quality data. Simultaneously, we incorporate consistency loss in the universal network to align the prediction of pancreatic EUS with tumor boundaries from other organs to mitigate the domain gap. To further reduce the cross-organ domain gap, the auxiliary network integrates multi-scale features from different organs, aiding the universal network in acquiring domain-invariant knowledge. Systematic experiments demonstrate that COTS-Nets significantly improves the accuracy of pancreatic cancer diagnosis. Additionally, we developed the Pancreatic Cancer Endoscopic Ultrasound (PCEUS) dataset, comprising 501 pathologically confirmed pancreatic EUS images, to facilitate model development.

Penghui Liu, Yi Niu, Xiaoxiong Zhong et al.

With the rapid evolution of the Industrial Internet of Things (IIoT), the boundaries and scale of the Internet are continuously expanding. Consequently, the limitations of traditional certificate-based Public Key Infrastructure (PKI) have become increasingly evident, particularly in scenarios requiring large-scale certificate storage, verification, and frequent transmission. These challenges are expected to be further amplified by the widespread adoption of post-quantum cryptography. In this paper, we propose a novel identity-based public key management framework for PKI based on post-quantum cryptography, termed \textit{IPK-pq}. This approach implements an identity key generation protocol leveraging NIST ML-DSA and random matrix theory. Building on the concept of the Composite Public Key (CPK), \textit{IPK-pq} addresses the linear collusion problem inherent in CPK through an enhanced identity mapping mechanism. Furthermore, it simplifies the verification of the declared public key's authenticity, effectively reducing the complexity associated with certificate-based key management. We also provide a formal security proof for \textit{IPK-pq}, covering both individual private key components and the composite private key. To validate our approach, formally, we directly implement and evaluate \textit{IPK-pq} within a typical PKI application scenario: Resource PKI (RPKI). Comparative experimental results demonstrate that an RPKI system based on \textit{IPK-pq} yields significant improvements in efficiency and scalability. These results validate the feasibility and rationality of \textit{IPK-pq}, positioning it as a strong candidate for next-generation RPKI systems capable of securely managing large-scale routing information.

Jinwu Hu, Zihao Lian, Zhiquan Wen et al.

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

Lang Nie, Yuan Mei, Kang Liao et al.

We present \textit{RopStitch}, an unsupervised deep image stitching framework with both robustness and naturalness. To ensure the robustness of \textit{RopStitch}, we propose to incorporate the universal prior of content perception into the image stitching model by a dual-branch architecture. It separately captures coarse and fine features and integrates them to achieve highly generalizable performance across diverse unseen real-world scenes. Concretely, the dual-branch model consists of a pretrained branch to capture semantically invariant representations and a learnable branch to extract fine-grained discriminative features, which are then merged into a whole by a controllable factor at the correlation level. Besides, considering that content alignment and structural preservation are often contradictory to each other, we propose a concept of virtual optimal planes to relieve this conflict. To this end, we model this problem as a process of estimating homography decomposition coefficients, and design an iterative coefficient predictor and minimal semantic distortion constraint to identify the optimal plane. This scheme is finally incorporated into \textit{RopStitch} by warping both views onto the optimal plane bidirectionally. Extensive experiments across various datasets demonstrate that \textit{RopStitch} significantly outperforms existing methods, particularly in scene robustness and content naturalness. The code is available at {\color{red}https://github.com/MmelodYy/RopStitch}.