论文检索

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.

Deep Learning · Generative Models and Autoencoders

Zhuguanyu Wu, Ruihao Gong, Yang Yong, Yushi Huang, Xiangyu Fan, Lei Yang, Dahua Lin, Xianglong Liu

Distribution Matching Distillation (DMD) is a widely used paradigm for accelerating inference in few-step video diffusion models. However, DMD-style training faces a structural bottleneck: the student-side auxiliary score network (the fake score) must closely track a continuously evolving generator. Updating the fake score too frequently increases training cost and can over-emphasize inner-loop tracking, while infrequent updates lead to tracking lag that destabilizes training and degrades generation consistency. To address this issue, we propose \textbf{Score Gradient Matching Distillation (SGMD)}. SGMD adopts a fake-score perspective by directly optimizing the fake score toward the teacher, while using teacher stop-gradient Fisher as a stable distribution-matching objective. We provide a gradient analysis that motivates this objective choice under ideal tracking. Building on this, SGMD introduces a pair of dual potentials: negative-residual (NR) for outer-loop correction and residual-contraction (RC) for inner-loop tracking. Empirically, compared to DMD, SGMD achieves an approximately $\sim 3\times$ training speedup and substantially improves motion dynamics for 4-step distilled models while preserving temporal consistency.

Applications · Computer Vision

Jueqi Liu, Xuechao Zou, Congyan Lang

Explicit surface reconstruction aims to recover high-fidelity meshes directly from point clouds. While existing methods achieve strong performance on scene-level data, they often rely on test-time optimization, resulting in a prohibitive runtime of several minutes. To address this bottleneck, we propose FastSESR, a two-stage framework for efficient scene-level explicit surface reconstruction. In the first stage, a lightweight triangular candidate network (TCN) captures local connections via an edge-factorized parameterization, enabling effective extraction of surface triangles from uniformly sampled points. In the second stage, an offset optimization network amortizes offset refinement into a small, fixed number of learnable update steps guided by TCN, producing geometries that are more suitable for triangulation. Experiments on multiple scene-level datasets show that FastSESR accelerates surface reconstruction by at least $20 \times$ over prior methods while maintaining competitive reconstruction quality. Moreover, evaluations on shape-level benchmarks indicate good generalization performance. Our code is available at https://anonymous.4open.science/r/FastSR-84C1.

Deep Learning · Generative Models and Autoencoders

Junchao Huang, Ziyang Ye, Xinting Hu, Tianyu He, Guiyu Zhang, Shaoshuai Shi, Jiang Bian, Li Jiang

Autoregressive video world models predict future visual observations conditioned on actions. While effective over short horizons, these models often struggle with long-horizon generation, as small prediction errors accumulate over time. Prior methods alleviate this by introducing pre-trained teacher models and sequence-level distribution matching, which incur additional computational cost and fail to prevent error propagation beyond the training horizon. In this work, we propose LIVE, a Long-horizon Interactive Video world modEl that enforces bounded error accumulation via a novel cycle-consistency objective, thereby eliminating the need for teacher-based distillation. Specifically, LIVE first performs a forward rollout from ground-truth frames and then applies a reverse generation process to reconstruct the initial state. The diffusion loss is subsequently computed on the reconstructed terminal state, providing an explicit constraint on long-horizon error propagation. Moreover, we provide an unified view that encompasses different approaches and introduce progressive training curriculum to stabilize training. Experiments demonstrate that LIVE achieves state-of-the-art performance on long-horizon benchmarks, generating stable, high-quality videos far beyond training rollout lengths.

General Machine Learning · Representation Learning

Ezgi Ozyilkan, Sharang Sriramu, Elza Erkip, Aaron Wagner, Jona Ballé

Neural compression is currently dominated by Nonlinear Transform Coding (NTC), which maps data to real-valued latents via continuous transforms. Despite its success, NTC suffers from train-test mismatch due to non-differentiable quantization, a ''smoothness bias'' inherent in continuous transforms that precludes optimality for certain sources, and a loss of ''shaping gain" due to the complexity of including high-dimensional vector quantization. We propose **SoftBinary Coding** (SBC), an end-to-end learning paradigm that bypasses these limitations by using a stochastic binary latent space. In the spirit of vector quantization, SBC employs discrete representations and compresses them through a novel fast binary channel simulation scheme, for which we provide a proof of rate optimality. Experimental gains on information-theoretic sources provide both theoretical and practical closure to NTC's limitations, establishing discrete binary structures as a viable path toward reaching optimal rate--distortion bounds. Surprisingly, SBC also achieves state-of-the-art performance on vector quantization of i.i.d. sources, exceeding Trellis Coded Quantization of the Gaussian source.

Deep Learning · Large Language Models

Bowen LIU, Zhi Wu, RunquanXie, Zhanhui Kang, Jia Li

Scaling verifiable training signals remains a key bottleneck for Reinforcement Learning from Verifiable Rewards (RLVR). Logical reasoning is a natural substrate: constraints are formal and answers are programmatically checkable. However, prior synthesis pipelines either depend on expert-written code or operate within fixed templates/skeletons, which limits growth largely to instance-level perturbations. We propose SSLogic, an agentic meta-synthesis framework that scales at the task-family level by iteratively synthesizing and refining executable Generator--Validator program pairs in a closed Generate--Validate--Refine loop, enabling continuous family evolution with controllable difficulty. To ensure reliability, we introduce a Multi-Gate Validation Protocol that combines multi-strategy consistency checks with Adversarial Blind Review, where independent agents must solve instances by writing and executing code to filter ambiguous or ill-posed tasks. Starting from 400 seed families, two evolution rounds expand to 953 families and 21,389 verifiable instances (from 5,718). Training on SSLogic-evolved data yields consistent gains over the seed baseline at matched training steps, improving SynLogic by +5.2, BBEH by +1.4, AIME25 by +3.0, and Brumo25 by +3.7. Code is available at https://anonymous.4open.science/r/Scaling-the-Scaling-Logic-6F4B/.

Deep Learning · Large Language Models

jialiang zhu, Gongrui Zhang, Xiaolong Ma, Lin Xu, Miaosen Zhang, Ruiqi Yang, Song Wang, Kai Qiu, Zhirong Wu, Qi Dai 等

LLM-based deep research agents are largely built on the ReAct framework. This linear design makes it difficult to revisit earlier states, branch into alternative search directions, or maintain global awareness under long contexts, often leading to local optima, redundant exploration, and inefficient search. We propose Re-TRAC, an agentic framework that performs cross-trajectory exploration by generating a structured state representation after each trajectory to summarize evidence, uncertainties, failures, and future plans, and conditioning subsequent trajectories on this state representation. This enables iterative reflection and globally informed planning, reframing research as a progressive process. Empirical results show that Re-TRAC consistently outperforms ReAct by 15–20% on BrowseComp with frontier LLMs. For smaller models, we introduce Re-TRAC-aware supervised fine-tuning, achieving state-of-the-art performance at comparable scales. Notably, Re-TRAC shows a monotonic reduction in tool calls and token usage across rounds, indicating progressively targeted exploration driven by cross-trajectory reflection rather than redundant search.

Deep Learning · Generative Models and Autoencoders

Alessandro Micheli, Yueqi Cao, Anthea Monod, Samir Bhatt

Computational optimal transport (OT) offers a principled framework for generative modeling. Neural OT methods, which use neural networks to learn an OT map (or potential) from data in an amortized way, can be evaluated out of sample after training, but existing approaches are tailored to Euclidean geometry. Extending neural OT to high-dimensional Riemannian manifolds remains an open challenge. In this paper, we prove that any method for OT on manifolds that produces discrete approximations of transport maps necessarily suffers from the curse of dimensionality: achieving a fixed accuracy requires a number of parameters that grows exponentially with the manifold dimension. Motivated by this limitation, we introduce Riemannian Neural OT (RNOT) maps, which are continuous neural-network parameterizations of OT maps on manifolds that avoid discretization and incorporate geometric structure by construction. Under mild regularity assumptions, we prove that RNOT maps approximate Riemannian OT maps with sub-exponential complexity in the dimension. Experiments on synthetic and real datasets demonstrate improved scalability and competitive performance relative to discretization-based baselines.

Deep Learning · Generative Models and Autoencoders

Ryien Hosseini, Pouya Gholami, Filippo Simini, Venkatram Vishwanath, Rebecca Willett, Henry (Hank) Hoffmann

We study the problem of generating structurally diverse graphs on $N$ unlabeled vertices. Given a space of such graphs $S_N$, a metric $d$, and a target cardinality $k$, the objective is to construct a set $\mathcal{G} \subset S_N$ that maximizes pairwise diversity under $d$. While neural generative models may appear appealing as a solution, standard approaches require samples from a target distribution that does not exist for dispersion problems. As a result, prior work is limited to brute-force combinatorial or iterative search methods. We instead treat diversity as an explicit optimization objective, an approach we term *Neural Graph Dispersion*. An ensemble of generators is optimized under a repulsive potential, producing diverse graphs by sampling along optimization trajectories as they disperse over $(S_N,d)$. Moreover, this approach allows us to generate an initial diverse graph set and, when desired, refine it under bespoke graph distances with minimal overhead. Extensive experiments show our method produces highly diverse graphs while scaling efficiently with respect to $N$ and $k$.

Optimization · Zero-order and Black-box Optimization

Alexander Chebykin, Tanja Alderliesten, Peter A.N Bosman

Hyperparameter Optimization (HPO) can lift the burden of tuning hyperparameters (HPs) of neural networks. HPO algorithms from the Population Based Training (PBT) family are efficient thanks to dynamically adjusting HPs every few steps of the weight optimization. Recent results indicate that the number of steps between HP updates is an important meta-HP of all PBT variants that can substantially affect their performance. Yet, no method or intuition is available for efficiently setting its value. We introduce Iterated Population Based Training (IPBT), a novel PBT variant that automatically adjusts this HP via restarts that reuse weight information in a task-agnostic way and leverage time-varying Bayesian optimization to reinitialize HPs. Evaluation on 8 image classification and reinforcement learning tasks shows that, on average, our algorithm matches or outperforms 5 previous PBT variants and other HPO algorithms (random search, ASHA, SMAC3), without requiring a budget increase or any changes to its HPs.

Deep Learning · Generative Models and Autoencoders

Xuhui Chen, Chao Long, Fei Hou, dongbo zhang, Shaohui Jiao, Wencheng Wang, Ying He

High-fidelity 3D generation remains difficult. Although some methods have proposed converting raw meshes to SDFs, it remains a lossy process. TripoSF presented a VAE training paradigm based on a rendering loss to circumvent this lossy SDF conversion, achieving high-precision surface reconstruction. However, because the rendering loss cannot supervise all the VAE outputs in the same way as SDF supervision, it limits detail and scalability.We present Focusing, a 3D VAE that improves efficiency by activating only the voxels that matter for a given view. Our key idea is a depth-driven voxel carving performed in the structured latent space: voxels inconsistent with the rendered depth are pruned before decoding. This concentrates learning on locally relevant geometry, reduces attention and decoding costs, and lowers video random access memory (VRAM) usage. To stabilize training and capture fine details, we further introduce an adaptive zooming strategy that adjusts camera intrinsics to keep the number of active voxels within a target range. The VAE is trained with a render-based loss on depth, normals, masks, and perceptual terms, and we add simple regularizers (e.g., sparse-voxel TV and a short warm-up with TSDF supervision) to reduce small holes and speed up convergence. Across standard reconstruction benchmarks, Focusing improves geometric accuracy (CD, F-score) over strong baselines while cutting VRAM consumption, which allows for training the resolution VAE on as little as 50GB of VRAM. These results show that local, view-consistent sparsity is an effective route to higher-resolution, more efficient 3D VAEs.

Applications · Robotics

Zhuoyang Liu, Jiaming Liu, Hao Chen, Jiale Yu, Ziyu Guo, Chengkai Hou, Xiangju Mi, Chenyang Gu, Renrui Zhang, Kun Wu 等

Vision-Language-Action (VLA) models have recently shown strong generalization, with some approaches seeking to explicitly generate linguistic reasoning traces or predict future observations prior to execution. However, explicit reasoning typically incurs non-negligible inference latency, which constrains the temporal resolution required for robotic manipulation. Moreover, such reasoning is confined to the linguistic space, imposing a representational bottleneck that struggles to faithfully capture ineffable physical attributes. To mitigate these limitations, we propose LaST$_0$, a framework that enables efficient reasoning before acting through a Latent Spatio-Temporal Chain-of-Thought (CoT), capturing fine-grained physical and robotic dynamics that are often difficult to verbalize. Specifically, we introduce a token-efficient latent CoT space that models future visual dynamics, 3D structural information, and robot proprioceptive states, and further extends these representations across time to enable temporally consistent implicit reasoning trajectories. Furthermore, LaST$_0$ adopts a dual-system architecture implemented via a Mixture-of-Transformers design, where a reasoning expert conducts low-frequency latent inference and an acting expert generates high-frequency actions conditioned on robotics-oriented latent representations. To facilitate coordination, LaST$_0$ is trained with heterogeneous operation frequencies, enabling adaptive switching during deployment. Across 10 real-world tasks spanning tabletop, mobile, and dexterous hand manipulation, LaST$_0$ improves mean success rates by 13%, 14% and 14% over prior SOTA VLA methods, respectively.

Deep Learning · Large Language Models

Hee Suk Yoon, Eunseop Yoon, Jaehyun Jang, SooHwan Eom, Ji Woo Hong, Mark Hasegawa-Johnson, Qi Dai, Chong Luo, Chang D. Yoo

While on-policy distillation offers dense supervision for training small reasoning models, its optimization dynamics in the multimodal domain remain under-explored. In this work, we challenge the standard monolithic view of Vision-Language Model (VLM) distillation by mathematically decomposing the loss into two distinct components: the language prior and visual grounding. Our analysis uncovers that gradient vectors for these components are nearly orthogonal, indicating that the objective of aligning with the teacher's language distribution is geometrically independent from the objective of matching its visual perception. Consequently, standard optimization passively follows a suboptimal compromise trajectory that implicitly balances the two objectives. Hypothesizing that visual grounding constitutes the primary bottleneck for vision-language reasoning, we introduce Visual Gradient Steering (VGS), a method that dynamically reorients the update vector to prioritize the visual subspace. Experimental results on multiple distillation settings and complex multimodal benchmarks demonstrate that VGS significantly outperforms the standard monolithic formulation of on-policy distillation, achieving superior grounding with minimal training overhead. Code will be released.

Applications · Computer Vision

Chendong Wang, Donglin Bai, Yifan Yang, Xiao Jin, Anlan Zhang, Rui Wang, Shiqi Jiang, Yuqing Yang, Hao Wu, Qi Dai 等

We present $\textit{Video-in-the-Loop}$ (ViTL), a two-stage long-video QA framework that preserves a fixed token budget by first $\textit{localizing}$ question-relevant interval(s) with a low-fps skim and then $\textit{answering}$ via span-aware reallocation of visual tokens at higher effective frame rate, emitting an interleaved output with both spans and the final option for direct attribution. We also introduce $\textit{VGrounding-QA}$, which converts description based event graphs into $\textit{span-grounded}$ multiple-choice QA by pairing each question with $\textit{ground-truth}$ time span(s) and related reasoning. ViTL is trained end-to-end with an interleaved group-relative objective that couples temporal IoU for localization with answer correctness, allowing credit to flow from answers back to spans without increasing compute. Under fixed token budgets, ViTL attains up to 8.6\% with 50\% less frame input on long-video QA and temporal grounding (e.g., Charades-STA, ActivityNet-Captions) and ablations show that span-aware token reallocation consistently surpasses uniform sampling. Together, $\textit{VGrounding-QA}$ and ViTL provide an interpretable, compute-efficient recipe for scalable long-video QA.