论文检索

Deep Learning · Attention Mechanisms

Yikang Yue, Yuqi Xue, Jian Huang

Long-context Large Language Model (LLM) inference has become the norm for today’s AI applications. However, it is severely bottlenecked by the increasing memory demands of its KV cache. Previous works have shown that self-speculative decoding with sparse attention, where tokens are drafted using a subset of the KV cache and verified in parallel with full KV cache, speeds up inference in a lossless way. However, this approach relies on standalone KV selection algorithms to select the KV entries used for drafting and overlooks that the criticality of each KV entry is inherently computed during verification. In this paper, we propose Vegas, a self-speculative decoding method with verification-guided sparse attention. Vegas identifies critical KV entries as a byproduct of verification and only loads these entries when drafting subsequent tokens. This not only improves draft token acceptance rate but also incurs low KV selection overhead, thereby improving decoding throughput. Vegas achieves 2.81× higher throughput over vanilla auto-regressive decoding and 1.29× improvement over state-of-the-art sparsity-based self-speculative decoding methods.

Reinforcement Learning · Multi-agent

Hu Fu, Pengyi Li, Hao Chen, Xuanyu Xiang, Biao Luo, Yihua Tan

Parameter Sharing (PS) is widely used to improve efficiency in Multi-Agent Reinforcement Learning (MARL), but it can limit behavioral diversity and degrade performance. This limitation stems from gradient conflicts among agents on shared weights, which hinders effective policy learning. To fully characterize this phenomenon, we propose Geometric Gradient Decomposition Analysis that decomposes gradients with respect to weight vector into radial (scale) and tangential (direction) components and uncover a key insight: agents largely agree on directional updates but substantially disagree on scale updates. Consequently, while recent methods split the shared network into agent-specific subnetworks to mitigate conflicts, they also discard shared directional updates, limiting training efficiency. To address this issue, we propose Hyperspherical Parameter Sharing (HPS), which explicitly decouples direction and scale in parameter sharing. Specifically, HPS constrains the shared backbone weights onto a Riemannian manifold(unit hypersphere), enforcing purely directional learning. Building on this, an agent-specific scale generator outputs multiplicative modulation factors to adjust each agent’s scales, thus preserving heterogeneous response magnitudes without disrupting the shared directions. Experiments on SMAC, SMACv2, VMAS and Predator Prey demonstrate that HPS effectively resolves the scale conflict, significantly outperforming state-of-the-art methods.

Deep Learning · Large Language Models

Shuofei Qiao, Yunxiang Wei, Xuehai Wang, Bin Wu, Boyang XUE, Ningyu Zhang, Hossein A. Rahmani, Wang Yanshan, Qiang Zhang, Keyan Ding 等

The rapid evolution of Large Language Models has catalyzed a surge in scientific idea production, yet this leap has not been accompanied by a matching advance in idea evaluation. The fundamental nature of scientific evaluation needs knowledgeable grounding, collective deliberation, and multi-criteria decision-making. However, existing idea evaluation methods often suffer from narrow knowledge horizons, flattened evaluation dimensions, and the inherent bias in LLM-as-a-Judge. To address these, we regard idea evaluation as a knowledge-grounded, multi-perspective reasoning problem and introduce **InnoEval**, a deep innovation evaluation framework designed to emulate human-level idea assessment. We apply a heterogeneous deep knowledge search engine that retrieves and grounds dynamic evidence from diverse online sources. We further achieve review consensus with an innovation review board containing reviewers with distinct academic backgrounds, enabling a multi-dimensional decoupled evaluation across multiple metrics. We construct comprehensive datasets derived from authoritative peer-reviewed submissions to benchmark InnoEval. Experiments demonstrate that InnoEval can consistently outperform baselines in point-wise, pair-wise, and group-wise evaluation tasks, exhibiting judgment patterns and consensus highly aligned with human experts.

Deep Learning · Large Language Models

Yuda Song, Lili Chen, Fahim Tajwar, REMI MUNOS, Deepak Pathak, J. Bagnell, Aarti Singh, Andrea Zanette

The success of RL for LLM post-training stems from an unreasonably uninformative source: a single bit of information per rollout as binary reward or preference label. At the other extreme, distillation offers dense supervision but requires demonstrations, which are costly and difficult to scale. We study natural language feedback as an intermediate signal: richer than scalar rewards, yet cheaper than complete demonstrations. Textual feedback is a natural mode of human interaction and is already abundant in many real-world settings, where users, tools, and automated judges routinely critique LLM outputs. Towards leveraging text feedback at scale, we formalize a multi-turn RL setup where text feedback is available during training but not at inference. Therefore, models must learn to internalize the feedback in order to improve their test-time single-turn performance. To do this, we propose two methods: Self Distillation, which trains the single-turn policy to match its own feedback-conditioned second-turn generations; and Feedback Modeling, which predicts the feedback as an auxiliary objective. We provide theoretical analysis on both methods, and empirically evaluate on reasoning puzzles, competition math, and creative writing tasks. Our results show that both methods consistently outperform strong baselines across benchmarks, highlighting the potential of RL with an additional source of rich supervision at scale.

Deep Learning · Large Language Models

Ao Li, Shangpeng Yang, Fahao Chen, Tianheng Xu, Peng Li, su zhou

Large Language Model (LLM)-based agents demonstrate strong reasoning and execution capabilities on complex tasks when guided by structured instructions, commonly referred to as workflows. However, existing workflow-assisted agent serving systems typically rely on predefined templates and shallow matching mechanisms, which limit their ability to capture deep semantic relationships and generalize to previously unseen tasks. To address these limitations, we propose a new workflow management paradigm that represents workflows using a unified graph, termed wGraph, where each node corresponds to an atomic operation. wGraph serves as a shared substrate from which task-specific workflows are dynamically instantiated. Building on wGraph primitives, we introduce GraphFlow, a system that efficiently integrates workflows into agent serving through two key designs. First, adaptive workflow generation dynamically constructs workflows from wGraph based on task semantics and constraint requirements. Second, workflow state management exploits wGraph structure to efficiently manage Key-Value (KV) caches, reducing redundant computation during agent serving. Extensive experiments across five benchmark datasets show that GraphFlow consistently outperforms state-of-the-art methods, yielding an average performance improvement of approximately 4.95 percentage points, while achieving an approximately 4× reduction in memory footprint.

Applications · Robotics

Zixuan Hu, Xuantuo Huang, Yancheng Li, Yichun Hu, Shengyong Xu, LINGYU DUAN

Navigating under non-stationary environment shifts poses a critical challenge for a Vision-and-Language Navigation (VLN) agent deployed in the wild. Yet, existing Test-Time Adaptation (TTA) methods for VLN largely treat online adaptation as transient, isolated updates, leading to catastrophic forgetting and negative transfer. To overcome these issues, we propose **I**nter-**D**omain Bridg**E** with Historical **A**ssets (**IDEA**), a novel TTA framework that transforms adaptation into the accumulation and composition of assets. Specifically, IDEA introduces soft prompts optimized via a Fisher-guided weighting scheme to capture the transferable knowledge. These optimized prompts are then augmented with domain coordinates to form a dynamic asset library. Leveraging this library, IDEA constructs a cross-domain bridge by projecting the target domain onto the convex hull of historical knowledge. These designs form a complementary loop: the evolving library underpins bridge construction, while the bridge provides superior initialization to accelerate asset optimization. Extensive experiments across REVERIE, R2R, and R2R-CE benchmarks demonstrate the consistent superiority of IDEA over existing methods, showcasing its ability to enable training-free adaptation via asset sharing. The code will be publicly available.

Deep Learning · Large Language Models

Ran Xu, Tianci Liu, Zihan Dong, Tony Yu, Ilgee Hong, Carl Yang, Linjun Zhang, Tuo Zhao, Haoyu Wang

Standard reward models typically predict scalar scores that fail to capture the multifaceted nature of response quality in non-verifiable domains, such as creative writing or open-ended instruction following. To address this limitation, we propose Rubric-ARM, a framework that jointly optimizes a rubric generator and a judge using reinforcement learning from preference feedback. Unlike existing methods that rely on static rubrics or disjoint training pipelines, our approach treats rubric generation as a latent action learned to maximize judgment accuracy. We introduce an alternating optimization strategy to mitigate the non-stationarity of simultaneous updates, providing theoretical analysis that demonstrates how this schedule reduces gradient variance during training. Extensive experiments show that Rubric-ARM achieves state-of-the-art performance among baselines on multiple benchmarks and significantly improves downstream policy alignment in both offline and online reinforcement learning settings.

General Machine Learning · Clustering

Yao Fu, Dong Hu, Zhi Wang

Multiview subspace clustering (MvSC) has shown remarkable potential in exploring underlying structures of high-dimensional data. However, existing MvSC methods still suffer from two shortcomings: (1) the commonly use of convex low-rank approximations inadequately capture high-order correlations across views, while sensitivity to noise and outliers degrades clustering performance, and (2) they lack the ability to preserve global correlations and local geometric patterns simultaneously. To address these issues, we propose a novel nonconvex regularized MvSC model with deep prior, which not only accurately characterizes the intrinsic low-rank structure and suppresses the effect of outliers, but also preserves local structural properties through deep networks. By mathematically analyzing the optimal solution of the optimization problem in our proposed model, we develop an efficient ADMM-based algorithm with provable convergence guarantees to solve it. Extensive experiments on various datasets demonstrate the superiority of the proposed model.

Deep Learning · Graph Neural Networks

Jiaming Zhuo, Shuai Zhai, Ziyi Ma, Kun Fu, Chuan Wang, Di Jin, Zhen Wang, Xiaochun Cao, Huazhu Fu, Liang Yang

Brain network classification is pivotal for diagnosing neurological disorders, yet clinical interpretability and the identification of discriminative biomarkers fundamentally rely on precise functional parcellation. However, existing graph learning models for brain network analysis typically suffer from a critical limitation termed spectral unreachability. This stems from the widely used coupled encoder-pooling architecture, where the inherent representation smoothing property of graph encoders (including Graph Neural Networks and Graph Transformers) inevitably corrupts the high-frequency topological signals essential for delineating sharp functional boundaries. To solve this issue, the Hierarchical Spectral Parcellation Network (HiSP-Net) is proposed. Adopting a *project-then-align* philosophy, HiSP-Net structurally decouples partition learning from representation smoothing. Specifically, this model is constructed as a hierarchy of Spectral Parcellation blocks. Within each block, node (or module) representations are mapped directly via a topology-agnostic projection into a partition space to preserve high-frequency details, while a Topology-Aware Alignment mechanism enforces spatial coherence using a joint structural objective. Extensive evaluations on real-world datasets show the capability of HiSP-Net in achieving superior classification performance and extracting interpretable functional biomarkers. The source code is publicly available at https://anonymous.4open.science/r/HiSP-Net-demo-0F62/

Social Aspects · Accountability, Transparency, and Interpretability

Ilya Lasy, Nora Cai, Kola Ayonrinde

Sparse Mixture of Experts (MoE) models scale more efficiently than dense models by routing tokens to modular expert networks that are only active when relevant to the task. A leading hypothesis for the performance of MoE models is that each expert specialises in a single, coherent domain. However, interpretability efforts that assume this hypothesis have generally been unsuccessful. We propose and present evidence for an alternative account that we call the *Superposed Specialisation Hypothesis* (SSH): experts specialise in a disjoint union of fine-grained features rather than one broad domain. Leveraging the SSH, we introduce *RouterInterp*, a method for interpreting expert routing that identifies Sparse Autoencoder features most predictive of routing decisions and produces unified natural language explanations. On gpt-oss-20b, explanations from RouterInterp predict expert routing with 77% higher accuracy than prior methods. This work provides a scalable method for generating concise and more accurate explanations of expert routing and increases our understanding of a previously uninterpretable component of foundation models.

Deep Learning · Graph Neural Networks

Fan Li, Xiaoyang Wang, Chen Chen, Wenjie Zhang

The increasing prevalence of large-scale hypergraphs poses significant computational challenges for hypergraph neural network (HNN) training. To address this, hypergraph condensation (HGC) distills large real hypergraphs into compact yet informative synthetic ones, going beyond traditional graph condensation (GC) methods limited to modeling pairwise relations. However, existing HGC methods rely on decoupled training architectures, where structure generators are pre-trained on the original hypergraph but not jointly optimized with condensed features during refinement, resulting in misaligned structures that degrade downstream utility. Moreover, trajectory-based optimization incurs substantial computational overhead in refinement, limiting condensation efficiency. To tackle these issues, we propose \textbf{A}nchor-guided \textbf{H}yper\textbf{G}raph \textbf{C}ondensation with \textbf{D}ual-level \textbf{D}iscrimination (\textbf{AHGCDD}), which consists of three key components: (1) a node initialization module based on Heat Kernel PageRank (HKPR) to encode structural knowledge into feature semantics; (2) an anchor-guided hyperedge synthesis scheme based on feature-level association for joint optimization of condensed features and structure; (3) a theoretically grounded dual-level discrimination objective for utility-preserving condensation without redundant HNN training. Extensive experiments demonstrate the superior effectiveness and efficiency of AHGCDD.

Deep Learning · Graph Neural Networks

Youda Mo, Chaobo He, Junwei Cheng, Peng Mei, Quanlong Guan

Dynamic graphs are pervasive in real-world systems, but their tightly entangled spatiotemporal evolution causes significant modeling challenges. Existing Dynamic Graph Neural Networks (DGNNs) lack a principled framework for systematically decoupling this multi-domain entanglement, raising two key problems: (i) representation drift caused by structural incompleteness, and (ii) signal distortion amplified by noise perturbation. These problems can accumulate over time, forming temporal redundancy that weakens robustness of DGNNs. In view of these, we propose DeR-Mamba(Decoupling for Robust Mamba), a multi-domain decoupling framework for robust DGNNs. To address (i), we develop the Multi-Particle Kernel Kalman observation field (MP-K$^2$alman), which achieves spatial decoupling by sampling latent evolution paths in kernel subspaces and performing Kalman-style updates to estimate structural states. To address (ii), we design the Adversarial-aware Frequency Decoupling Module (AFDM), which performs frequency-domain decoupling and dynamic cross-frequency modulation to purify spectral signals. Finally, a self-consistent dynamic graph state-space system performs temporal decoupling to control redundancy, suppressing residual disturbances through discretized cross-time modeling and selective snapshot scanning. Extensive experiments on benchmark datasets with adversarial attacks validate its superior robustness.

Deep Learning · Graph Neural Networks

Xu Bai, Bin Lu, kunzhang, Shengbo Chen, Xinbing Wang, Chenghu Zhou, Meng Jin

Graph coarsening is a graph dimensionality reduction technique that aims to construct a smaller and more tractable graph while preserving the essential structural and semantic properties of the original graph. However, most existing methods rely on pair-wise similarity matching, where each node independently searches for its best partner based on global information. This \textit{selfishness} matching paradigm incurs substantial computational and memory overhead. To address this problem, we shift to a \textit{non-selfishness} principle that prioritizes the collective interference of neighborhood in coarsening, and propose an efficient method named \texttt{NOPE}, which achieves linear memory consumption and near-linear computational complexity in the number of nodes. Furthermore, we derive a faster variant \texttt{NOPE}*, which reduces $\mathcal{O}(\Delta\cdot d)$ interference evaluation to $\mathcal{O}(d)$ based on the local isotropy assumption, and consequently alleviates the computational bottleneck for high-degree nodes. Experimental results show that \texttt{NOPE}* achieves $1.8–10\times$ speedup over \texttt{NOPE} and surpass almost all baselines with 1-3 orders of magnitude acceleration. Meanwhile, learning on coarsened graphs yields comparable performance to original graphs, and can even show superior performance over LLM-based graph reasoning owing to compact graph information. The code can be available at https://anonymous.4open.science/r/NOPE-FA74.

Deep Learning · Large Language Models

Jiakuan Xie, Pengfei Cao, Kang Liu, Jun Zhao

Despite the advancing reasoning capabilities of large language models (LLMs), many reasoning evaluations are increasingly compromised by data contamination, which induces unreliable **contaminated reasoning** on leaked inputs. While this phenomenon is widely observed, its underlying mechanism remains poorly understood, hindering the ability to distinguish generalization from memorization and to develop effective solutions. In this work, we first identify a distinctive signal of contaminated reasoning, namely the mutual information decay between representations and gradients. Our mechanistic analysis reveals that contaminated models exhibit pronounced eigenspectrum concentration in their representations, leading to a low-dimensional computation regime. Under leaked inputs, this mechanism weakens the linear coupling between representations and gradients, manifested as a structural decay of the singular values in the whitened space. We show that this narrowing geometry mathematically implies a reduction in mutual information, and further demonstrate the practical utility of our analysis by successfully restoring the reasoning behavior of contaminated models, achieving an 11.03% improvement in average consistency with the base model over the strongest baseline.

Deep Learning · Graph Neural Networks

Liang Yang, Shuai Zhai, Ziyi Ma, Jiaming Zhuo, Di Jin, Chuan Wang, Zhen Wang, Xiaochun Cao

The construction of the brain functional network often follows the hand-crafted Correlation Coefficients of blood-oxygen-level-dependent (BOLD) time series without any learnable components. Meanwhile, most efforts are made to the models, such as graph neural networks, that make predictions with the constructed brain network as input. Unfortunately, the fixed brain network may lose critical information during construction and lead to difficulty in performance improvement, even with deliberately designed graph models. From this perspective, the current situation is similar to the machine learning community, i.e., hand-crafted features and learnable predictors, before the advent of representation learning. In fact, the brain network can be regarded as a graph-structured learnable representation of the brain. By drawing on representation learning, this paper presents the Brain Representation (BRep) learning problem. To this end, the widely used linear and nonlinear correlations are enhanced to be high-order, parametric, and learnable. The flexible brain representation makes the following predictor simple and leads the framework to possess an end-to-end characteristic. The framework is implemented by combining the parametric correlation and a TopK sparsification. Theoretical analysis guarantees the model's universal approximation to any U/V-statistics. Extensive evaluations demonstrate that the proposed BRep possesses superior performance, high efficiency, and interpretability. The source code is publicly available at https://anonymous.4open.science/r/BRep-demo-1A3E/

Deep Learning · Algorithms

Keigo Nishida, Eren Mehmet KIRAL, Kenichi Bannai, Mohammad Emtiyaz Khan, Thomas Moellenhoff

We propose a new algorithm enabling stable training under low-precision computations. We call this algorithm Log-normal Multiplicative Dynamics (LMD), and derive it by taking inspiration from the robustness of biological neural networks. Such networks use synapses whose size follow log-normal distribution and whose fluctuations follow noisy multiplicative dynamics. There has been no scalable algorithm to train modern deep networks that incorporates all such synaptic properties. We are able to include them in LMD by using a variational formulation where a log-normal posterior distribution is used. We show several results that involve low-precision matrix multiplications in the forward passes. This includes results for training Vision Transformer and GPT-2 from scratch. Our findings suggest that biologically inspired multiplicative dynamics offer a promising direction for future energy-efficient hardware.

Deep Learning · Graph Neural Networks

Haoyu Chen, Jinsong Wang, Zening Zhao, Kai Shi, Zongpu Wei, Jianhao Li

Federated Graph Learning (FGL) enables distributed clients to collaboratively train graph neural networks while strictly preserving data privacy.However, existing FGL methods implicitly assume the reliability of local graph structures and lack elastic awareness of parameter importance during model aggregation, leading to representation degradation under topological noise and catastrophic forgetting caused by model drift. To address these challenges,we propose GraphP-FL, a general personalized FGL framework.(1)we design a self-supervised dynamic topology reconstruction mechanism on the client side. This mechanism mines implicit dependencies to adaptively rectify noisy topologies, effectively suppressing topological noise propagation and capturing precise structural relationships for high-quality representations.(2)we introduce a Fisher-based Elastic Parameter Alignment (FRPA) algorithm. FRPA imposes anisotropic regularization constraints in the parameter space to precisely quantify parameter importance, enabling the model to strictly preserve critical local knowledge while flexibly aligning with the global model, thus effectively overcoming catastrophic forgetting.Extensive experiments on seven benchmarks (including biochemical molecules, social networks, and large-scale encrypted traffic) demonstrate that GraphP-FL significantly outperforms state-of-the-art methods, improving accuracy by up to 8.6% while exhibiting superior generalization and robustness.

Applications · Computer Vision

Zirui Pan, Xin Wang, Yipeng Zhang, Yuwei Zhou, Wenwu Zhu

Despite rapid advances in text-to-video generation, state-of-the-art generative models still suffer from producing temporally incoherent and unrealistic motion for videos. The key weakness of existing works is that they commonly treat videos as frame sequences and directly adopt Flow Matching objectives, which are originally designed for images. This practice fails to explicitly model motion priors or temporal dependencies, resulting in suboptimal dynamics that may appear incoherent and unrealistic. To solve this problem, we propose Temporal-aware Flow Matching (TFM), a novel training paradigm that embeds inter-frame constraints into the flow objective, leading to temporally coherent motion modeling in video generation. More specifically, the proposed TFM enforces temporal correlations across frames while retaining the desirable properties of Flow Matching, and further introduces a residual-type loss that aligns naturally with this new flow. We theoretically prove that models trained with TFM are able to exhibit remarkably enhanced temporal perception ability and better capture motion dynamics. Notably, TFM imposes no additional cost during inference and is applicable to any model using Flow Matching. Extensive experiments demonstrate that our TFM can significantly improve motion realism across diverse motion types. Generated videos are presented at https://tfm-2026.github.io.

Deep Learning · Generative Models and Autoencoders

Haoran Fang, Jinjie Fang, Tianxing Man, Wanli Shi, Xingchen Li, Bin Gu

Diffusion Transformers (DiTs) have achieved state-of-the-art generative performance, yet their iterative denoising process remains computationally expensive and energy-intensive. Spiking Neural Networks (SNNs) offer a promising neuromorphic alternative for energy efficiency; however, the non-differentiable nature of spiking neurons makes direct training difficult, positioning ANN-to-SNN conversion as a more practical, training-free solution. In this paper, we identify a critical challenge unique to converting DiTs: standard fixed-scale spiking neurons fail to accommodate the highly dynamic activation ranges inherent across denoising steps. This mismatch leads to cumulative errors that significantly degrade generation fidelity. To resolve this, we propose a novel conversion framework featuring Multi-Threshold (MT) neurons and a Membrane Potential Error-Feedback (MPEF) mechanism. MT neurons expand the expressive capacity of discrete spikes by employing a multi-level firing strategy. Concurrently, MPEF exploits the temporal correlation between successive denoising steps to recycle residual membrane potential, effectively compensating for information loss and mitigating distribution shifts without retraining. Extensive experiments on ImageNet demonstrate that our framework achieves competitive generative quality with superior energy efficiency, establishing a new performance benchmark for spiking Diffusion Transformers.

Hao Ren, Zetong Bi, Yiming Zeng, Le Zheng, Zhi Li, Zhaoliang Wan, Lu Qi, Hui Cheng

Diffusion models are effective for waypoint prediction in visual navigation, but standard sampling and test time guidance can produce unsafe or inefficient trajectories when updates drift off the training manifold. We propose Fisher Preserving Guidance with Outer Product Span Projection, a training-free inference method that avoids large Fisher drift associated with off-distribution actions while optimizing a task objective. Our method computes the Fisher-preserving update via a low-rank Jacobian factorization, requiring only a single backward pass per step and enabling real-time use. We further introduce Truncated Fisher Denoising Sensitivity as an uncertainty signal and use it for robust multi-sample action blending. Experiments on toy and realistic navigation benchmarks, including Maze2D with TSDF-based guidance, PushT with official Diffusion Policy weights, and visual navigation in simulation and on real robots, demonstrate consistent improvements in performance over strong diffusion-policy baselines without additional training.