Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by integrating up-to-date external knowledge, yet real-world web environments present unique challenges. These limitations manifest as two key challenges: pervasive misinformation in the web environment, which introduces unreliable or misleading content that can degrade retrieval accuracy, and the underutilization of web tools, which, if effectively employed, could enhance query precision and help mitigate this noise, ultimately improving retrieval results in RAG systems. To address these issues, we propose WebFilter, a novel RAG framework that generates source-restricted queries and filters out unreliable content. This approach combines a retrieval filtering mechanism with a behavior- and outcome-driven reward strategy, optimizing both query formulation and retrieval outcomes. Extensive experiments demonstrate that WebFilter improves answer quality and retrieval precision, outperforming existing RAG methods on both in-domain and out-of-domain benchmarks.

Multiple clustering aims to discover diverse latent structures from different perspectives, yet existing methods generate exhaustive clusterings without discerning user interest, necessitating laborious manual screening. Current multi-modal solutions suffer from static semantic rigidity: predefined candidate words fail to adapt to dataset-specific concepts, and fixed fusion strategies ignore evolving feature interactions. To overcome these limitations, we propose Multi-DProxy, a novel multi-modal dynamic proxy learning framework that leverages cross-modal alignment through learnable textual proxies. Multi-DProxy introduces 1) gated cross-modal fusion that synthesizes discriminative joint representations by adaptively modeling feature interactions. 2) dual-constraint proxy optimization where user interest constraints enforce semantic consistency with domain concepts while concept constraints employ hard example mining to enhance cluster discrimination. 3) dynamic candidate management that refines textual proxies through iterative clustering feedback. Therefore, Multi-DProxy not only effectively captures a user's interest through proxies but also enables the identification of relevant clusterings with greater precision. Extensive experiments demonstrate state-of-the-art performance with significant improvements over existing methods across a broad set of multi-clustering benchmarks.

Jinze Li, Yang Zhang, Xin Yang, Jiayi QU, Jinfeng Xu, Shuo Yang, Junhua Ding, Edith Cheuk-Han Ngai. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Offline batch inference is gaining popularity as a cost-effective solution for latency-insensitive tasks, such as model evaluation and data curation. As the latency objective is highly relaxed, maximizing throughput is the primary goal in offline inference. Previous studies focused solely on throughput optimization within a batch. However, the diverse resource demands (compute-intensive vs. memory-intensive) across a wide range of applications make these approaches less effective, as imbalanced resource demands between batches restrict optimization opportunities.

The explosion of multimedia data in information-rich environments has intensified the challenges of personalized content discovery, positioning recommendation systems as an essential form of passive data management. Multimodal sequential recommendation, which leverages diverse item information such as text and images, has shown great promise in enriching item representations and deepening the understanding of user interests. However, most existing models rely on heuristic fusion strategies that fail to capture the dynamic and context-sensitive nature of user-modal interactions. In real-world scenarios, user preferences for modalities vary not only across individuals but also within the same user across different items or categories. Moreover, the synergistic effects between modalities-where combined signals trigger user interest in ways isolated modalities cannot-remain largely underexplored. To this end, we propose CAMMSR, a Category-guided Attentive Mixture of Experts model for Multimodal Sequential Recommendation. At its core, CAMMSR introduces a category-guided attentive mixture of experts (CAMoE) module, which learns specialized item representations from multiple perspectives and explicitly models inter-modal synergies. This component dynamically allocates modality weights guided by an auxiliary category prediction task, enabling adaptive fusion of multimodal signals. Additionally, we design a modality swap contrastive learning task to enhance cross-modal representation alignment through sequence-level augmentation. Extensive experiments on four public datasets demonstrate that CAMMSR consistently outperforms state-of-the-art baselines, validating its effectiveness in achieving adaptive, synergistic, and user-centric multimodal sequential recommendation.

The ''pre-training, prompt-tuning'' has emerged as a pivotal paradigm in advancing the performance of graph representation learning models across a wide range of downstream tasks. This paradigm leverages the power of pre-trained models and task-specific prompts to bridge the gap between general graph representations and task-specific requirements. Early graph prompt tuning approaches relied on task-specific designs for Graph Neural Networks (GNNs), limiting their adaptability across diverse pre-training strategies. In contrast, another promising line of research has investigated universal graph prompt tuning, which operates directly in the input graph's feature space and builds a theoretical foundation that universal graph prompt tuning can theoretically achieve an equivalent effect of any prompting function, eliminating dependence on specific pre-training strategies. Recent works propose selective node-based graph prompt tuning to pursue more ideal prompts. However, we argue that selective node-based graph prompt tuning inevitably compromises the theoretical foundation of universal graph prompt tuning. In this paper, we strengthen the theoretical foundation of universal graph prompt tuning by introducing stricter constraints, demonstrating that adding prompts to all nodes is a necessary condition for achieving the universality of graph prompts. To this end, we propose a novel model and paradigm, Learning and Editing Universal GrAph Prompt Tuning (LEAP), which preserves the theoretical foundation of universal graph prompt tuning while pursuing more ideal prompts. Specifically, we first build the basic universal graph prompts to preserve the theoretical foundation and then employ actor-critic reinforcement learning to select nodes and edit prompts. Extensive experiments on graph- and node-level tasks across various pre-training strategies in both full-shot and few-shot scenarios show that LEAP consistently outperforms fine-tuning and other prompt-based approaches.

Although existing multimodal recommendation models have shown promising performance, their effectiveness continues to be limited by the pervasive data sparsity problem. This problem arises because users typically interact with only a small subset of available items, leading existing models to arbitrarily treat unobserved items as negative samples. To this end, we propose VI-MMRec, a model-agnostic and training cost-free framework that enriches sparse user-item interactions via similarity-aware virtual user-item interactions. These virtual interactions are constructed based on modality-specific feature similarities of user-interacted items. Specifically, VI-MMRec introduces two different strategies: (1) Overlay, which independently aggregates modality-specific similarities to preserve modality-specific user preferences, and (2) Synergistic, which holistically fuses cross-modal similarities to capture complementary user preferences. To ensure high-quality augmentation, we design a statistically informed weight allocation mechanism that adaptively assigns weights to virtual user-item interactions based on dataset-specific modality relevance. As a plug-and-play framework, VI-MMRec seamlessly integrates with existing models to enhance their performance without modifying their core architecture. Its flexibility allows it to be easily incorporated into various existing models, maximizing performance with minimal implementation effort. Moreover, VI-MMRec introduces no additional overhead during training, making it significantly advantageous for practical deployment. Comprehensive experiments conducted on six real-world datasets using seven state-of-the-art multimodal recommendation models validate the effectiveness of our VI-MMRec.

Recent advancements in multimodal recommendations, which leverage diverse modality information to mitigate data sparsity and improve recommendation accuracy, have gained significant attention. However, existing multimodal recommendations overlook the critical role of user representation initialization. Unlike items, which are naturally associated with rich modality information, users lack such inherent information. Consequently, item representations initialized based on meaningful modality information and user representations initialized randomly exhibit a significant semantic gap.

Constructing radio maps traditionally requires extensive site surveys with precise location labels, resulting in costly and time-consuming calibration. Conventional approaches derive labels from inertial measurement units (IMUs), but are constrained by device-level access permissions and the need for pre-installed IMU hardware. In this paper, we present a calibration-free radio map construction method that relies solely on Channel State Information (CSI) measurements, thereby obviating the need for location labels. Our key insight is to embed CSI data into a two-dimensional geographic space using a neural network, without any label information. The primary challenge is to preserve the real-world topology in the embedded space. To address this issue, we propose a novelTopology-Guided Manifold Learning (TGML)approach that learns a low-dimensional embedding through self-supervision based on its mapping to a topological map in the geographic space. Specifically, we introduce an embedding network that learns locally smoothed proximity, thus creating a compact low-dimensional representation in the latent space. We further employ a regularized transport method with a differentiable Sinkhorn distance to establish an optimal mapping between the latent and geographic spaces. We use the resulting mapping deviation to supervise the training of the embedding network, enabling continuous refinement through self-supervision. Experiments in an office environment show that TGML achieves an average localization error of 2.37m and a relative beam estimation error of 6.68%, outperforming state-of-the-art methods.

The exponential growth of online information has made it increasingly difficult for users to identify valuable and relevant content. Recommender systems have emerged as a critical solution to this challenge by tailoring content to individual preferences. With the proliferation of diverse multimedia services, human interaction with the digital world has become inherently multimodal. Consequently, recommender systems capable of comprehending and interpreting multimodal information can more effectively align with individual preferences. With its recent surge in research attention, the field of multimodal recommender systems (MRS) still lacks a comprehensive technical survey. Existing surveys suffer from two critical limitations: 1) Insufficient technical depth: Prior works predominantly focus on categorizing and discussing general structure, neglecting rigorous technical analysis of methodologies and architectures. 2) Absence of cutting-edge works: Due to the rapid evolution of AI technologies, current surveys fail to discuss the latest works that adopt the most advanced techniques. To bridge these gaps, this survey conducts a systematic and technical review of advanced MRS works from MRS's inception to the present. We organize existing works into coherent taxonomies based on their structure and provide in-depth analyses of methodological innovations at each component, including Feature Extraction, Encoder, Multimodal Fusion, and Loss Function. Moreover, we further discuss potential future directions for developing and enhancing MRS. This survey serves as technical guidance for researchers and practitioners, offering insights into the developments, techniques, and future directions of MRS. Notably, we maintain a public repository:https://github.com/Jinfeng-Xu/Awesome-Multimodal-Recommender-Systems.

Beyond traditional user recommendation, group recommendation is a new and popular task that provides recommendations for a group of users. Previous works aggregate member preferences in the group to infer group preference, but this often leads to a coarse-grained inference for group preferences limited by users’ individual preferences. To this end, we exploit that user preferences can be inferred and refined by exploring the group preferences that they participated in. These refined preferences offer additional information beyond the original individual preferences, enabling more fine-grained and satisfactory group preference inference. In this work, we propose a novel Dual-Granularity Graph Variational Auto-Encoder framework (DGGVAE) for group recommendation, which jointly reveals group preferences from both coarse granularity and fine granularity to comprehensively learn group preferences. Specifically, we design a Group Preference Extractor module that extracts group preferences from these two granularities: coarse granularity, which is revealed through original member preferences, and fine granularity, which is revealed through refined member preferences. To extract the correlation between groups, a Group Representation Enhancement module is proposed, which enhances group representations by information from the most similar groups. However, the coarse- and fine-grained group preferences contain uncertainty due to the gap between the original and refined member preferences. To better incorporate dual-granularity group preferences, we design granularity-specific graph variational encoders that learn Gaussian variables on the semantic information for each group. Moreover, with the conditional independence assumption, the granularity-specific Gaussian node embeddings are fused according to the generalized product-of-experts (gPoE), where the semantic information in each granularity is weighted based on the estimated uncertainty level. Extensive experiments show the superiority of DGGVAE over various state-of-the-art methods in training efficiency and accuracy on both group and user recommendation tasks.

Learning-based Android malware detectors degrade over time due to natural distribution drift caused by malware variants and new families. This paper systematically investigates the challenges classifiers trained with empirical risk minimization (ERM) face against such distribution shifts and attributes their shortcomings to their inability to learnstablediscriminative features. Invariant learning theory offers a promising solution by encouraging models to generate stable representations across environments that expose the instability of the training set. However, the lack of prior environment labels, the diversity of drift factors, and low-quality representations caused by diverse families make this task challenging. To address these issues, we propose TIF, the first temporal invariant training framework for malware detection, which aims to enhance the ability of detectors to learn stable representations across time. TIF organizes environments based on application observation dates to reveal temporal drift, integrating specialized multi-proxy contrastive learning and invariant gradient alignment to generate and align environments with high-quality, stable representations. TIF can be seamlessly integrated into any learning-based detector. Experiments on a decade-long dataset show that TIF excels, particularly in early deployment stages, addressing real-world needs and outperforming state-of-the-art methods.

As multimedia information proliferates, multimodal recommendation systems have garnered significant attention. These systems leverage multimodal information to alleviate the data sparsity issue inherent in recommendation systems, thereby enhancing the accuracy of recommendations. Due to the natural semantic disparities among multimodal features, recent research has primarily focused on cross-modal alignment using self-supervised learning to bridge these gaps. However, aligning different modal features might result in the loss of valuable interaction information, distancing them from ID embeddings. It is crucial to recognize that the primary goal of multimodal recommendation is to predict user preferences, not merely to understand multimodal content. To this end, we propose a new Multi-level sElf-supervised learNing for mulTimOdal Recommendation (MENTOR) method, which effectively reduces the gap among modalities while retaining interaction information. Specifically, MENTOR begins by extracting representations from each modality using both heterogeneous user-item and homogeneous item-item graphs. It then employs a multilevel cross-modal alignment task, guided by ID embeddings, to align modalities across multiple levels while retaining historical interaction information. To balance effectiveness and efficiency, we further propose an optional general feature enhancement task that bolsters the general features from both structure and feature perspectives, thus enhancing the robustness of our model.

Graph Collaborative Filtering (GCF) has emerged as a dominant paradigm in modern recommendation systems, excelling at modeling complex user-item interactions and capturing high-order collaborative signals. Most existing GCF models predominantly rely on simplified graph architectures like LightGCN, which strategically remove feature transformation and activation functions from vanilla graph convolution networks. Through systematic analysis, we reveal that feature transformation in message propagation can enhance model representation, though at the cost of increased training difficulty. To this end, we propose FourierKAN-GCF, a novel framework that adopts Fourier Kolmogorov-Arnold Networks as efficient transformation modules within graph propagation layers. This design enhances model representation while decreasing training difficulty. Our FourierKAN-GCF can achieve higher recommendation performance than most widely used GCF backbone models and can be integrated into existing advanced self-supervised models as a backbone, replacing their original backbone to achieve enhanced performance. Extensive experiments on three public datasets demonstrate the superiority of FourierKAN-GCF.

Modern recommender systems face critical challenges in handling information overload while addressing the inherent limitations of multimodal representation learning. Existing methods suffer from three fundamental limitations: (1) restricted ability to represent rich multimodal features through a single representation, (2) existing linear modality fusion strategies ignore the deep nonlinear correlations between modalities, and (3) static optimization methods failing to dynamically mitigate the over-smoothing problem in graph convolutional network (GCN). To overcome these limitations, we propose HPMRec, a novel Hypercomplex Prompt-aware Multimodal Recommendation framework, which utilizes hypercomplex embeddings in the form of multi-components to enhance the representation diversity of multimodal features. HPMRec adopts the hypercomplex multiplication to naturally establish nonlinear cross-modality interactions to bridge semantic gaps, which is beneficial to explore the cross-modality features. HPMRec also introduces the prompt-aware compensation mechanism to aid the misalignment between components and modality-specific features loss, and this mechanism fundamentally alleviates the over-smoothing problem. It further designs self-supervised learning tasks that enhance representation diversity and align different modalities. Extensive experiments on four public datasets show that HPMRec achieves state-of-the-art recommendation performance.

The proliferation of the Internet of Things (IoT) is reshaping daily life while generating vast amounts of distributed user data, raising critical security and privacy concerns. Although federated learning-based intrusion detection systems (FL-based IDS) have shown promise in detecting potential attacks in such distributed environments, non-independent and identically distributed (non-i.i.d.) real-world applications’ data lead to considerable performance degradation. Current schemes mitigate this issue by sharing subsets of local clients’ data to guide consistency among clients; however, it inevitably compromises data privacy. In response to these challenges, we propose FedCKD, a federated framework that integrates well-designed Contrastive Knowledge Distillation for network intrusion detection. By utilizing a teacher-student model architecture on client devices, FedCKD aligns the feature maps of various models, ensuring consistent feature representation across clients without any lack of raw data. Furthermore, the incorporation of cluster contrastive learning enhances the separation between anomalous and normal instances, improving the model’s discriminative capabilities. We conduct extensive experiments on the NSL-KDD and UNSW-NB15 datasets, demonstrating that FedCKD outperforms existing methods, achieving over 90% f1-score (an increase of 3.67%). scenarios. Our approach effectively alleviates the impact of data heterogeneity while preserving data privacy, highlighting FedCKD as a robust and privacy-preserving solution for intrusion detection in IoT networks.

Leveraging long contexts is crucial for advanced AI systems, but attention computation poses a scalability challenge. While scaled dot-product attention (SDPA) exhibits token sparsity, i.e. only a few pivotal tokens significantly contribute to output, exploiting this sparsity remains challenging. Existing methods either suffer from quality degradation or require substantial additional resources. We show that identifying pivotal tokens is a Maximum Inner Product Search (MIPS) problem. However, existing MIPS solutions are not well-suited for SDPA, as they are not GPU-friendly and often underperform due to the separated query and key distributions. This paper introduces HashAttention, framing pivotal token identification as a recommendation problem. Given a query, HashAttention encodes keys and queries in Hamming space, capturing the required semantic similarity, using learned mapping functions. HashAttention efficiently identifies pivotal tokens for a given query using bitwise operations and computes attention using only these tokens, improving the overall attention efficiency. Trained on generic data, HashAttention reduces tokens used by up to $16\times$ with minimal quality loss, requiring only 32 bits of auxiliary memory per token. Sparsity can be further improved to $32\times$ through task-specific fine-tuning. On A100 GPU, at $32\times$ sparsity, incorporating HashAttention reduces attention latency by up to $4.3\times$ in GPT-FAST and $2.54\times$ in FlashDecode, and achieves up to $3.12\times$ higher throughput for GPT-FAST.

Diffusion Transformers (DiTs) dominate video generation but their high computational cost severely limits real-world applicability, usually requiring tens of minutes to generate a few seconds of video even on high-performance GPUs. This inefficiency primarily arises from the quadratic computational complexity of 3D full attention with respect to the context length. In this paper, we propose a training-free framework termed Sparse VideoGen (SVG) that leverages the inherent sparsity in 3D full attention to boost inference efficiency. We reveal that the attention heads can be dynamically classified into two groups depending on distinct sparse patterns: (1) Spatial Head, where only spatially-related tokens within each frame dominate the attention output, and (2) Temporal Head, where only temporally-related tokens across different frames dominate. Based on this insight, SVG proposes an online profiling strategy to capture the dynamic sparse patterns and predicts the type of attention head. Combined with a novel hardware-efficient tensor layout transformation and customized kernel implementations, SVG achieves up to 2.28$\times$ and 2.33$\times$ end-to-end speedup on CogVideoX-v1.5 and HunyuanVideo, respectively, while preserving generation quality. Our code will be open-sourced upon publication.

The data sparsity problem significantly hinders the performance of recommender systems, as traditional models rely on limited historical interactions to learn user preferences and item properties. While incorporating multimodal information can explicitly represent these preferences and properties, existing works often use it only as side information, failing to fully leverage its potential. In this paper, we propose MDVT, a model-agnostic approach that constructs multimodal-driven virtual triplets to provide valuable supervision signals, effectively mitigating the data sparsity problem in multimodal recommendation systems. To ensure high-quality virtual triplets, we introduce three tailored warm-up threshold strategies: static, dynamic, and hybrid. The static warm-up threshold strategy exhaustively searches for the optimal number of warm-up epochs but is time-consuming and computationally intensive. The dynamic warm-up threshold strategy adjusts the warm-up period based on loss trends, improving efficiency but potentially missing optimal performance. The hybrid strategy combines both, using the dynamic strategy to find the approximate optimal number of warm-up epochs and then refining it with the static strategy in a narrow hyper-parameter space. Once the warm-up threshold is satisfied, the virtual triplets are used for joint model optimization by our enhanced pair-wise loss function without causing significant gradient skew. Extensive experiments on multiple real-world datasets demonstrate that integrating MDVT into advanced multimodal recommendation models effectively alleviates the data sparsity problem and improves recommendation performance, particularly in sparse data scenarios.

The efficiency and scalability of graph convolution networks (GCNs) in training recommender systems remain critical challenges, hindering their practical deployment in real-world scenarios. In the multimodal recommendation (MMRec) field, training GCNs requires more expensive time and space costs and exacerbates the gap between different modalities, resulting in sub-optimal recommendation accuracy. This paper critically points out the inherent challenges associated with adopting GCNs during the training phase in MMRec, revealing that GCNs inevitably create unhelpful and even harmful pairs during model optimization and isolate different modalities. To this end, we propose FastMMRec, a highly efficient multimodal recommendation framework that deploys graph convolutions exclusively during the testing phase, bypassing their use in training. We demonstrate that adopting GCNs solely in the testing phase significantly improves the model's efficiency and scalability while alleviating the modality isolation problem often caused by using GCNs during the training phase. We conduct extensive experiments on three public datasets, consistently demonstrating the performance superiority of FastMMRec over competitive baselines while achieving efficiency and scalability.

Large Language Models (LLMs) hold significant potential for advancing fact-checking by leveraging their capabilities in reasoning, evidence retrieval, and explanation generation. However, existing benchmarks fail to comprehensively evaluate LLMs and Multimodal Large Language Models (MLLMs) in realistic misinformation scenarios. To bridge this gap, we introduce RealFactBench, a comprehensive benchmark designed to assess the fact-checking capabilities of LLMs and MLLMs across diverse real-world tasks, including Knowledge Validation, Rumor Detection, and Event Verification. RealFactBench consists of 6K high-quality claims drawn from authoritative sources, encompassing multimodal content and diverse domains. Our evaluation framework further introduces the Unknown Rate (UnR) metric, enabling a more nuanced assessment of models' ability to handle uncertainty and balance between over-conservatism and over-confidence. Extensive experiments on 7 representative LLMs and 4 MLLMs reveal their limitations in real-world fact-checking and offer valuable insights for further research. RealFactBench is publicly available at https://github.com/kalendsyang/RealFactBench.git.

Video generation models have rapidly progressed, positioning themselves as video world models capable of supporting decision-making applications like robotics and autonomous driving. However, current benchmarks fail to rigorously evaluate these claims, focusing only on general video quality, ignoring important factors to world models such as physics adherence.To bridge this gap, we propose WorldModelBench, a benchmark designed to evaluate the world modeling capabilities of video generation models in application-driven domains. WorldModelBench offers two key advantages: (1) Against to nuanced world modeling violations: By incorporating instruction-following and physics-adherence dimensions, WorldModelBench detects subtle violations, such as irregular changes in object size that breach the mass conservation law—issues overlooked by prior benchmarks. (2) Aligned with large-scale human preferences: We crowd-source 67K human labels to accurately measure 14 frontier models. Using our high-quality human labels, we further fine-tune an accurate judger to automate the evaluation procedure, achieving 9.9% lower error in predicting world modeling violations than GPT-4o with 2B parameters. In addition, we demonstrate that training to align human annotations by maximizing the rewards from the judger noticeably improve the world modeling capability. The dataset is hosted in HuggingFace at https://huggingface.co/datasets/Efficient-Large-Model/worldmodelbench. The code to run evaluation is available at https://github.com/WorldModelBench-Team/WorldModelBench.

Leveraging attention sparsity to accelerate long-context large language models (LLMs) has been of great importance recently. However, most existing sparse attention algorithms use a fixed budget of how many tokens to use in their computations. This simple static decision raises critical issues in real-world deployment because it fails to account for the dynamic nature of real-world scenarios, where the optimal balance between accuracy and efficiency can vary greatly. In this paper, we reveal a key insight that leveraging the idea of top-$p$ sampling (a.k.a., nucleus sampling) in sparse attention could enable efficient and adaptive budget decisions. Based on this, we propose Twilight, a framework that enhances any existing sparse attention algorithm with adaptive budget decision capabilities without sacrificing accuracy. Empirical results show that Twilight can adaptively prune up to 98% tokens with nearly no accuracy loss in both mid- and long-context scenarios, leading to a $1.4\times$ speedup over state-of-the-art sparse attention mechanisms.

Diffusion Transformers (DiTs) are essential for video generation but suffer from significant latency due to the quadratic complexity of attention. By computing only critical tokens, sparse attention reduces computational costs and offers a promising acceleration approach. However, we identify that existing methods fail to approach optimal generation quality under the same computation budget for two reasons: (1) Inaccurate critical token identification: current methods cluster tokens based on position rather than semantics, leading to imprecise aggregated representations. (2) Excessive computation waste: critical tokens are scattered among non-critical ones, leading to wasted computation on GPUs, which are optimized for processing contiguous tokens. In this paper, we propose SVG2, a training-free framework that maximizes identification accuracy and minimizes computation waste, achieving a Pareto frontier trade-off between generation quality and efficiency. The core of SVG2 is semantic-aware permutation, which clusters and reorders tokens based on semantic similarity using k-means. This approach ensures both a precise cluster representation, improving identification accuracy, and a densified layout of critical tokens, enabling efficient computation without padding. Additionally, SVG2 integrates Top-p dynamic budget control and customized kernel implementations, achieving up to $2.30\times$ and $1.89\times$ speedup while maintaining a PSNR of up to $30$ and $26$ on HunyuanVideo and Wan 2.1, respectively. Our code is open-sourced at https://github.com/svg-project/Sparse-VideoGen.

Multimodal recommender systems utilize a variety of information types to model user preferences and item properties, aiding in the discovery of items that align with user interests. Rich multimodal information alleviates inherent challenges in recommendation systems, such as data sparsity and cold start problems. However, multimodal information further introduces challenges in terms of robustness and generalization capability. Regarding robustness, multimodal information magnifies the risks associated with information adjustment and inherent noise, posing severe challenges to the stability of recommendation models. For generalization capability, multimodal recommender systems are more complex and difficult to train, making it harder for models to handle data beyond the training set, posing significant challenges to model generalization capability. In this paper, we analyze the shortcomings of existing robustness and generalization capability enhancement strategies in the multimodal recommendation field. We propose a sharpness-aware minimization strategy focused on batch data (BSAM), which effectively enhances the robustness and generalization capability of multimodal recommender systems without requiring extensive hyper-parameter tuning. Furthermore, we introduce a mixed loss variant strategy (BSAM+), which accelerates convergence and achieves remarkable performance improvement. We provide rigorous theoretical proofs and conduct experiments with nine advanced models on five widely used datasets to validate the superiority of our strategies. Moreover, our strategies can be integrated with existing robust training and data augmentation strategies to achieve further improvement, providing a superior training paradigm for multimodal recommendations.

Encrypted traffic classification is essential for network security and management. However, the encrypted nature makes it challenging to extract representative features from raw traffic data. Existing end-to-end methods ignore byte correlations within packets and potential correlations among packets, hindering the learning of real traffic semantics and leading to suboptimal performance. This paper proposes MsETC, a multi-scale contrastive attention representation learning method for encrypted traffic classification. MsETC divides the raw packet byte sequence into multi-scale patches and then extracts dual views for contrastive learning from both the inter-patch and intra-patch perspectives. This allows the model to capture correlations among bytes within a packet as well as the potential interactions between packets. Extensive experiments on real-world datasets demonstrate that the proposed method achieves superior classification performance with lower complexity.

Group activities are important behaviors in human society, providing personalized recommendations for groups is referred to as the group recommendation task. Existing methods can usually be categorized into two strategies to infer group preferences: 1) determining group preferences by aggregating members' personalized preferences, and 2) inferring group consensus by capturing group members' coherent decisions after common compromises. However, the former would suffer from the lack of group-level considerations, and the latter overlooks the fine-grained preferences of individual users. To this end, we propose a novel group recommendation method AlignGroup, which focuses on both group consensus and individual preferences of group members to infer the group decision-making. Specifically, AlignGroup explores group consensus through a well-designed hypergraph neural network that efficiently learns intra- and inter-group relationships. Moreover, AlignGroup innovatively utilizes a self-supervised alignment task to capture fine-grained group decision-making by aligning the group consensus with members' common preferences. Extensive experiments on two real-world datasets validate that our AlignGroup outperforms the state-of-the-art on both the group recommendation task and the user recommendation task, as well as outperforms the efficiency of most baselines.

The deployment of learning-based models to detect malicious activities in network traffic flows is significantly challenged by concept drift. With evolving attack technology and dynamic attack behaviors, the underlying data distribution of recently arrived traffic flows deviates from historical empirical distributions over time. Existing approaches depend on a significant amount of labeled drifting samples to facilitate the deep model to handle concept drift, which faces labor-intensive manual labeling and the risk of label noise. In this paper, we propose ReCDA, a Concept Drift Adaptation method with Representation enhancement, which consists of a self-supervised representation enhancement stage and a weakly-supervised classifier tuning stage. Specifically, in the initial stage, ReCDA introduces drift-aware perturbation and representation alignment to facilitate the model in acquiring robust representations from drift-aware and drift-invariant perspectives. Moreover, in the subsequent stage, a meticulously crafted instructive sampling strategy and a robust representation constraint encourage the model to learn discriminative knowledge about benign and malicious activities during fine-tuning, thereby enhancing performance further. We conduct comprehensive evaluations on several benchmark datasets under varying degrees of concept drift. The experiment results demonstrate the superior adaptability and robustness of the proposed method.

Vehicular Ad Hoc Network(VANET) is a hot topic, how to design an efficient authentication scheme is the core problem in this field. With the considerable progress in computing capability and side-channel attack capability, some seemingly safe third parties have also become less safe. Our paper focuses on weakening the use of trusted authority(TA) and improving the fault tolerance of authentication schemes. At first, all private keys are generated by users. Next, the ephemeral Diffie-Hellman key exchange scheme is used to construct a PID authentication. Moreover, we build a partial key delivery method without a secure channel. In the batch message authentication part, we use the Shamir secret share method to build a ($t,\ n$) threshold signature scheme that guarantees batch signatures cannot be tampered with fewer than$t$people involved. Finally, we make a comparison of time cost and communication cost with the other four schemes. Our scheme has a remarkable advantage over other schemes by comparing experimental data.

Network technology developments are blurring the security boundary of the Internet of Things (IoT), making it difficult for traditional border-based security architectures to cope with endless internal attacks. The Zero Trust Architecture (ZTA), with its core concept of “Never Trust, Always Verify”, effectively addresses this problem. Continuous Authentication (CA) is an indispensable component of the zero trust IoT. However, existing CA protocols are impractical to deploy on zero trust IoT due to their dependence on specific properties and thirst for resources. Therefore, this paper proposes a continuous authentication protocol, namely STCA, that uses stacked tokens to ensure the legitimacy of entities during a whole session. Through the theoretical analysis and simulation results, STCA resists several common attacks. The comparison analysis indicates that STCA has a better performance compared with related CA protocols, thereby demonstrating it is more suitable for zero trust IoT.

Nowadays, Content Delivery Network (CDN) is widely used for its convenience in providing services. However, the increasing demand for bandwidth puts tremendous pressure on CDN. Inspired by the great potential of periodic broadcasting to save bandwidth, we suggest an Efficient and Adaptive Content Delivery System (EACDS) to decrease traffic and shorten video content delivery delays. Specifically, we propose the Peak cutting and Valley filling for VBR (PVV) and the Exhaustive Periodic Broadcasting algorithm (EPB) to process videos and arrange slices with low bandwidth and delay. Furthermore, we introduce an enhanced version of EPB that supports Fast-Forwarding, namely FFB. To effectively deal with the constantly changing Internet, we also demonstrate an adaptive decision model that switches distribution schemes according to the online user scale. Extensive experiments show that the EACDS is excellent in many aspects. The PVV and decision model can save up to 57% of bandwidth and 54.7% of traffic respectively. Compared with some existing schemes, EPB allows the most negligible delay with the same bandwidth, and FFB supports fast-forwarding in periodic broadcasting.

The real-world data tends to be heavily imbalanced and severely skew the data-driven deep neural networks, which makes Long-Tailed Recognition (LTR) a massive challenging task. Existing LTR methods seldom train Vision Transformers (ViTs) with Long-Tailed (LT) data, while the off-the-shelf pretrain weight of ViTs always leads to unfair comparisons. In this paper, we systematically investigate the ViTs' performance in LTR and propose LiVT to train ViTs from scratch only with LT data. With the observation that ViTs suffer more severe LTR problems, we conduct Masked Generative Pretraining (MGP) to learn generalized features. With ample and solid evidence, we show that MGP is more robust than supervised manners. Although Binary Cross Entropy (BCE) loss performs well with ViTs, it struggles on the LTR tasks. We further propose the balanced BCE to ameliorate it with strong theoretical groundings. Specially, we derive the unbiased extension of Sigmoid and compensate extra logit margins for deploying it. Our Bal-BCE contributes to the quick convergence of ViTs in just a few epochs. Extensive experiments demonstrate that with MGP and Bal-BCE, LiVT successfully trains ViTs well without any additional data and outperforms comparable state-of-the-art methods significantly, e.g., our ViT-B achieves 81.0% Top-1 accuracy in iNaturalist 2018 without bells and whistles. Code is available at https://github.com/XuZhengzhuo/LiVT.

The large-scale heterogeneous network (LS-HetNet) provides seamless interconnection of everything by integrating different networks into a unified one, but it also comes with a significant risk of insider attacks that are difficult to address through traditional border-based security measures. Trust and reputation management (TRM) is a practical approach to solving this problem. However, past schemes are struggled to cope with the cross-domain architecture, scalability, and context dependency required by LS-HetNet, and have faced potential trust attacks. Hence, this paper proposes an efficient and robust cross-domain trust reputation model, MBC-Trust, to meet the above requirements simultaneously. MBC-Trust adopts a multi-level blockchain framework and a cloud-fog computing paradigm to support frequent and concurrent cross-domain trust sharing. Numerous network entities can access trust in an efficient and scalable manner through the collaboration of local and monitoring chains. In addition, a well-designed time decay function, recommendation trust factors, and other measures enhance the robustness of the model to attacks during trust evaluation and provide sensitive context awareness. Experiments demonstrate that MBC-Trust is highly efficient, robust, and scalable.

In the real world, data tends to follow long-tailed distributions w.r.t. class or attribution, motivating the challenging Long-Tailed Recognition (LTR) problem. In this paper, we revisit recent LTR methods with promising Vision Transformers (ViT). We figure out that 1) ViT is hard to train with longtailed data. 2) ViT learns generalized features in an unsupervised manner, like mask generative training, either on longtailed or balanced datasets. Hence, we propose to adopt unsupervised learning to utilize long-tailed data. Furthermore, we propose the Predictive Distribution Calibration (PDC) as a novel metric for LTR, where the model tends to simply classify inputs into common classes. Our PDC can measure the model calibration of predictive preferences quantitatively. On this basis, we find many LTR approaches alleviate it slightly, despite the accuracy improvement. Extensive experiments on benchmark datasets validate that PDC reflects the model’s predictive preference precisely, which is consistent with the visualization.

Network Intrusion Detection (NID) works as a kernel technology for the security network environment, obtaining extensive research and application. Despite enormous efforts by researchers, NID still faces challenges in deploying on resource-constrained devices. To improve detection accuracy while reducing computational costs and model storage simultaneously, we propose a lightweight intrusion detection approach based on self-knowledge distillation, namely LNet-SKD, which achieves the trade-off between accuracy and efficiency. Specifically, we carefully design the DeepMax block to extract compact representation efficiently and construct the LNet by stacking DeepMax blocks. Furthermore, considering compensating for performance degradation caused by the lightweight network, we adopt batchwise self-knowledge distillation to provide the regularization of training consistency. Experiments on benchmark datasets demonstrate the effectiveness of our proposed LNet-SKD, which outperforms existing state-of-the-art techniques with fewer parameters and lower computation loads.

Deep learning-based fine-grained network intrusion detection systems (NIDS) enable different attacks to be responded to in a fast and targeted manner with the help of large-scale labels. However, the cost of labeling causes insufficient labeled samples. Also, the real fine-grained traffic shows a long-tailed distribution with great class imbalance. These two problems often appear simultaneously, posing serious challenges to fine-grained NIDS. In this work, we propose a novel semi-supervised fine-grained intrusion detection framework, SF-IDS, to achieve attack classification in the label-limited and highly class imbalanced case. We design a self-training backbone model called RI-1DCNN to boost the feature extraction by reconstructing the input samples into a multichannel image format. The uncertainty of the generated pseudo-labels is evaluated and used as a reference for pseudo-label filtering in combination with the prediction probability. To mitigate the effects of fine-grained class imbalance, we propose a hybrid loss function combining supervised contrastive loss and multi-weighted classification loss to obtain more compact intra-class features and clearer interclass intervals. Experiments show that the proposed SF-IDS achieves 3.01% and 2.71% Marco-F1 improvement on two classical datasets with 1% labeled, respectively.

Fog computing facilitates low-latency computing and nearby storage of device data in the Industrial Internet of Things (IIoT). Frequent internal interactions between fog nodes can support complex industrial processes, creating a secure fog interaction environment imperative. However, decentralized, mobile, and heterogeneous fog nodes make static cryptography-based security schemes challenging to cope with internal attacks. Adding a trust management model (TMM) to the fog layer is an effective solution to address this issue. Existing fog-oriented TMMs mostly ignore the inconsistent behavior of nodes or blindly trust fake recommendations from highly trusted nodes, which makes TMMs vulnerable to trust attacks from individuals or groups. Also, diverse industrial scenarios require TMMs to sensitively capture changes in fog node behaviors and contexts, rather than relying solely on binary ratings. This paper proposes a reliable distributed trust management model, TFog, to secure industrial IoT fog node interactions. It provides all well-defined model components and their collaboration methods to operate independently on fog nodes in a scalable form. Novel recommendation filtering algorithms, bi-directional interaction rating, and multi-party trust update strategies are proposed to resist trust attacks. Experiments show that TFog has good sensitivity and convergence, also can effectively resist eight typical trust attacks.

Fault tolerance is one of the major design goals for HPC. The emergence of non-volatile memories (NVM) provides a solution to build fault tolerant HPC. Data in NVM-based main memory are not lost when the system crashes because of the non-volatility nature of NVM. However, because of volatile caches, data must be logged and explicitly flushed from caches into NVM to ensure consistence and correctness before crashes, which can cause large runtime overhead. In this paper, we introduce an algorithm-based method to establish crash consistence in NVM for HPC applications. We slightly extend application data structures or sparsely flush cache blocks, which introduce ignorable runtime overhead. Such extension or cache flushing allows us to use algorithm knowledge to reason data consistence or correct inconsistent data when the application crashes. We demonstrate the effectiveness of our method for three algorithms, including an iterative solver, dense matrix multiplication, and Monte-Carlo simulation. Based on comprehensive performance evaluation on a variety of test environments, we demonstrate that our approach has very small runtime overhead (at most 8.2% and less than 3% in most cases), much smaller than that of traditional checkpoint, while having the same or less recomputation cost.