arXiv subjects

Kun He

Publications and source records attributed to Kun He.

At least 19 recordsLinked to original sources

arXiv2026

The Cost of Mismatch: Noise Amplification in Zeroth-Order Reinforcement Learning

Zeroth-order (ZO) optimization has emerged as a promising approach for memory-efficient large language model (LLM) post-training, yet its reliability in reinforcement learning with verifiable rewards (RLVR) remains largely unexplored. Motivated by the distinct data regimes of on-policy RL and fixed-data optimization, we develop a generic variance-theoretic framework to investigate what limits the reliability of ZO gradient estimation. Our analysis reveals that estimation reliability is structurally determined by whether the two finite-difference evaluations are performed on matched data. On-policy reinforcement learning (RL) under the mismatched-data regime produces a $1/μ^2$ oracle-noise amplification that sets an irreducible noise floor as the perturbation scale $μ$ decreases, whereas Boltzmann Targeted SFT (BOLT) under the matched-data regime eliminates this amplification via noise cancellation. We further derive convergence guarantees with explicit constants that cleanly separate the two regimes. These results motivate the ZO-BOLT algorithm, which builds on BOLT's matched-data advantage for reliable ZO optimization of an RLVR-aligned objective. Controlled perturbation-scale sweeps and matched-versus-mismatched end-to-end comparisons validate the predicted effect of data matching. Experiments across three models and two mathematical reasoning benchmarks further show that ZO-BOLT consistently improves over the pretrained baseline while retaining a near-inference memory footprint. The code is publicly available at https://github.com/lianmin-chen/the_cost_of_mismatch.

cs.LG↗

arXiv2026

ProAlign: Progressive Positional and Prototype-Guided Alignment for Aerial-Ground Person Re-Identification

Aerial-Ground Person Re-Identification (AG-ReID) aims to match pedestrian images captured by unmanned aerial vehicles (UAVs) and ground-based surveillance cameras. The task remains highly challenging due to severe viewpoint discrepancies, frequent occlusions, and substantial domain gaps between aerial and ground imagery. Beyond these observable factors, existing methods often overlook two fundamental structural issues: cumulative positional drift during hierarchical feature extraction and semantic inconsistency across cross-view feature distributions. To address these challenges, we propose ProAlign, a Progressive Positional and Prototype-Guided Alignment Network for AG-ReID. ProAlign comprises two key components: a Layer-wise Progressive Positional Embedding (LPPE) module and a View-Aware Prototype Contrastive Learning (PCL) module. LPPE performs adaptive positional calibration throughout transformer layers by employing conditional positional encoding in shallow layers to mitigate local spatial distortions, while introducing learnable static positional embeddings in deeper layers to reinforce global semantic priors. Meanwhile, PCL maintains view-specific aerial and ground prototypes for each identity and enforces cross-view semantic consistency via a dual-prototype contrastive objective. Extensive experiments on CARGO and two real-world AG-ReID benchmarks demonstrate the effectiveness of ProAlign, achieving state-of-the-art overall retrieval performance on CARGO with improvements of 1.60% in Rank-1 accuracy and 5.01% in mAP over the strongest competing methods.

cs.CV↗

arXiv2026

Fast Almost-Uniform Sampling of Random $k$-SAT Solutions

We study approximately uniform sampling of satisfying assignments from random $k$-SAT formulas. Fix a constant $B>0$. For every $k\ge 2^{20}$ and density $0<α\le 2^k/k^{16}$, we prove that, with high probability over a random formula on $n$ variables, every accuracy $n^{-B}\le\varepsilon\le 1/2$ admits a sampler whose output distribution is within total variation distance $\varepsilon$ of the uniform distribution on satisfying assignments with expected running time $$\widetilde O_B\left(\mathrm{poly}(k,α+1)n^{2+η}\right),\quad 0<η\le 2^{16-k}.$$ In particular, the exponent of $n$ is $2+O(2^{-k})$ while the dependence on the width and density is polynomial with a universal degree. This improves the parameter-dependent polynomial exponent of the sampling algorithm of Chen, Lonkar, Wang, Yang, and Yin (STOC 2025) in the density regime $2^k/\mathrm{poly}(k)$. Our algorithm separates a high-degree core from the remaining variables. Every residual formula on the good variables is a deletion subformula of a common bounded-dependency formula. Adapting the rejection--repair framework of Liu, Wang, Yin, Zhang and Zhou (arXiv 2026), we construct an executable sampler for all such residual formulas using ordinary connected violation sets replacing weighted 2-tree witnesses and use it to simulate approximate block heat-bath updates on the core. To analyze the resulting core chain, we integrate out the good variables, represent the core marginal by an absolutely convergent polymer expansion and combine a component-swap variance comparison with an adaptive exploration bound. Absorbing single-component interactions into the local weights leaves interactions of cluster size at least two; retaining their additional decay gives a spectral gap of at least $1/(2s n^{2^{12-k}})$ where $s$ counts core components. Together with the residual sampler, this yields the stated running time.

cs.DS↗

arXiv2026

SRA: Selective Residual Adaptation for LLM Reasoning Enhancement

Knowledge distillation transfers reasoning capabilities from large language models to lightweight students. However, distilled small models still struggle with multi-step reasoning. We observe that the distribution divergences between teacher and student models along reasoning paths are highly sparse, and target tokens almost always fall within the student top-$K$ candidate set. Based on this finding, unlike approaches updating student backbones or requiring runtime teacher computation, we propose Selective Residual Adaptation (SRA), a plug-and-play framework that internalizes localized corrections. Concretely, SRA employs a Discrepancy Gate to identify divergence positions alongside a Candidate Residual Adapter for top-$K$ logit calibration to correct the mismatch. Theoretically, we establish error bounds ensuring strict sequence level performance gains. Across mathematical benchmarks on Qwen3 models from 0.6B to 4B, SRA consistently enhances distilled student models, raising average Pass@16 by up to 4.37 percentage points while preserving general capabilities.

cs.LG↗

arXiv2026

Reinforcement Learning-driven Diffusion Models: Language, Vision, and Beyond

Diffusion models (DMs) have steadily demonstrated outstanding performance in high-fidelity, controllable generation across diverse scenarios, following the "pretraining-finetuning" paradigm established in LLMs. Despite these advances, conventional DMs still face critical limitations in sophisticated reasoning, autonomous exploration of generation pathways, and adaptability to complex, human-aligned objectives. Inspired by RLHF's success in LLMs, RL has recently emerged as a pivotal post-training mechanism to iteratively refine DMs through reward-driven trajectory optimization. This paper presents the first comprehensive review of RL-driven DMs and an empirical analysis of how RL reshapes the diffusion process towards experience-centric, goal-directed generation. We propose a novel technical panorama elucidating RL's multifaceted roles in diffusion-based language, multimodal, robotic, and scientific models, with a granular focus on diffusion trajectory optimization and reward engineering. Through extensive experiments on multiple task-specific benchmarks, we systematically evaluate representative methods to intuitively highlight the advantages of different RL post-training strategies. Furthermore, we reveal and dissect three crucial bottlenecks that currently hinder the real-world deployment of RL-driven DMs: reward design fragility, exploration-exploitation trade-off in high-dimensional generation spaces, and training instability under sparse rewards. Based on these perspectives, we provide a principled roadmap for the integration of RL into DMs and outline promising future research directions for robust, interpretable RL-diffusion systems. We also construct a public repository to continuously track the latest developments in this rapidly evolving field: https://github.com/MAIR-Lab-HUST/Awesome-RL-for-Diffusion-Models.

cs.AI↗

arXiv2026

RocketAgent: A Long-Horizon Engineering Agent for Multidisciplinary Design of Liquid-Rocket Thrust Chambers

Liquid-rocket thrust-chamber design involves interdependent analyses in which downstream constraints can require earlier design decisions to be revisited. Managing these dependencies across heterogeneous tools requires consistent design information and coordinated updates throughout the workflow. We present RocketAgent, a long-horizon engineering agent for multidisciplinary preliminary design of liquid-rocket thrust chambers. A single plan-owning Coding Agent coordinates engineering skills for performance sizing, subsystem optimization, geometry generation, and multiphysics assessment. A provenance-aware knowledge graph supports method selection, while a typed Design Intermediate Representation maintains shared parameters, artifacts, and decisions. Revision-aware checks invalidate affected results and block superseded inputs, with consequential changes subject to engineering approval. In a representative simulation-based design, RocketAgent continued from an infeasible cooling search through an engineer-authorized operating-point revision, identified feasible subsystem designs, and coordinated subsequent geometry generation and multiphysics assessment to support final configuration selection. Separate module tests assessed surrogate predictions and nozzle adaptation. A two-configuration comparison across three controlled scenarios verified the expected dependency invalidations and superseded-input blocking before solver execution. The representative case demonstrates sustained coordination across a multidisciplinary design workflow, while the controlled tests establish the behavior of the revision mechanisms supporting that execution.

cs.AI↗

arXiv2026

ITO: Multi-View Alignment and Training-Time Fusion for Image-Text Pretraining

Image--text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield representations that remain partially organized by modality rather than by semantics. We propose ITO, a framework addressing this limitation through two complementary mechanisms with distinct roles. Multimodal multiple alignment enriches supervision by constructing diverse cross-modal correspondences from multi-view image augmentations, providing the primary source of discriminative gain. A lightweight training-time multimodal fusion module then acts as a training-time regularizer, encouraging the encoders to produce features that are compatible under fusion. Crucially, the fusion module is discarded at inference, preserving the efficiency of standard dual-encoder architectures while incurring training-time overhead only. Extensive experiments across pretraining scales from millions to billions of image--text pairs show that ITO outperforms strong contrastive baselines at moderate scale and yields consistent gains over CLIP under identical data, backbone, and compute at the 100M--1B scale, on classification, retrieval, and multimodal benchmarks. Controlled comparisons show that these gains are not explained by additional training compute or image augmentation alone. Our analysis further reveals that the contribution of fusion grows with data scale and that it stabilizes optimization, mitigating the late-stage overfitting commonly observed in aggressive contrastive learning. Code is available at https://github.com/showstarpro/ITO.

cs.CV↗

arXiv2026

Rethinking Contrastive Loss in CLIP Post-training: A Complementary Framework with Frozen Text Encoder

CLIP serves as a foundational vision-language model and the de facto vision encoder for downstream VLMs such as LLaVA. Post-training offers a lightweight route to refine CLIP, but recent work argues that the standard contrastive loss is unsuitable for post-training due to catastrophic forgetting under small batches, motivating designs that abandon the contrastive objective in favor of distillation. We revisit this premise and find that, for the InfoNCE objective, the reported forgetting is driven primarily not by insufficient negatives but by an inappropriate magnitude of the contrastive temperature $τ$: with $τ$ set sufficiently small, contrastive post-training improves rather than degrades the pretrained CLIP, which we explain through the temperature dependence of the InfoNCE gradient. Building on this finding, we propose \textbf{ComCLIP}, a lightweight single-epoch post-training recipe that freezes CLIP's text encoder---so the refined vision encoder is a drop-in replacement with unchanged architecture and inference cost---and trains the vision encoder with a properly-tempered contrastive loss, an MSE anchoring loss against the original CLIP, and a relational distillation loss from DINOv2. Over multiple seeds, ComCLIP matches the self-distillation baseline CLIP-Refine on zero-shot classification while significantly improving the transferability of visual features, measured by linear probing ($48.99$ vs.\ $42.28$ on ViT-B/16), and on ViT-L/14 it also improves MMVP over CLIP-Refine ($24.20$ vs.\ $19.01$); CLIP-Refine remains stronger on image-text retrieval. Used as a drop-in vision encoder for LLaVA-1.5-7B without re-aligning the projector or LLM, ComCLIP yields no net change across $8$ VLM benchmarks, i.e., the refinement does not break downstream compatibility. Code and models are available at https://github.com/showstarpro/ComCLIP.git.

cs.CV↗

arXiv2026

Lollypop: Camera-to-Motion-Capture Calibration Verification with a Reference Target

Camera-to-motion-capture (mocap) calibration is essential for using mocap as ground truth in robotics, AR/VR, and other computer vision tasks. However, the calibration can drift after deployment, while calibration residuals and visual inspection provide limited independent verification. We present Lollypop, a fiducial-mocap reference target for independent calibration verification. The target couples an ArUco fiducial with a mocap marker constellation so the visual center and tracked centroid represent the same physical point. Given a candidate calibration, verification projects the mocap point into the image and measures its disagreement with the detected fiducial center. Experiments show sub-pixel nominal error, sensitivity to controlled extrinsic perturbations, and increasing error during an illustrative mixed-handling sequence.

cs.CV↗

arXiv2026

SDDBMs: Soft Denoising Diffusion Bridge Models

Diffusion bridge models leverage Doob's $h$-transform to construct stochastic transports between arbitrary endpoint distributions, and have shown strong potential in image-to-image translation and restoration. However, most existing bridge models rely on hard endpoint conditioning, which forces the terminal state to match a prescribed target exactly. This hard constraint induces terminal-boundary singularities: the terminal law collapses to a Dirac measure, and the resulting drift coefficients become ill-conditioned near the endpoint. In this paper, we propose Soft Denoising Diffusion Bridge Models (SDDBMs), a generalized framework that regularizes diffusion bridges directly at the level of their terminal constraints. Instead of imposing an exact endpoint, SDDBMs prescribe a non-degenerate Gaussian terminal marginal under the transformed path measure, with a flexible terminal center and variance. Starting from this prescribed marginal, we develop a complete closed-form construction of the soft bridge, including the Gaussian terminal reweighting and soft $h$-function, the induced Gaussian forward marginals and $\mathbf{x}_0$-free dynamics. Theoretically, SDDBMs provide a unified probabilistic perspective that encompasses existing diffusion bridge models, including DDBMs, GOUB, and UniDB, as special cases under specific parameter choices. Extensive experiments on image restoration tasks demonstrate that SDDBMs achieve improved numerical stability and superior generation quality over existing bridge-based methods.

cs.AI↗

arXiv2026

Compressing Value Predictions for Learning-Augmented Metrical Task Systems

Learning-augmented algorithms for metrical task systems (MTS) can exploit predictions of canonical dual values, but existing formulations typically require a prediction for every state. We study whether these predictions can be compressed to a small set of representative states while retaining their algorithmic value. We introduce landmark-compressed value predictions, in which the predictor reports predicted dual values only at $m$ landmarks and the remaining values are reconstructed by a Lipschitz extension. Our algorithm achieves additive excess cost $O(T\,r(L) + \sum_t δ_t)$, where $r(L)$ is the covering radius of the landmarks and $δ_t$ measures prediction error up to additive shifts; local and value-dependent bounds refine this guarantee. For sparse landmark sets on unit-spaced finite lines, we prove a matching $Ω(T r_m)$ lower bound for every randomized algorithm using fixed landmarks, even with advance access to their entire exact absolute-value table. The prediction interface also matters: on two states with one landmark, exact absolute values permit horizon-independent excess, whereas exact relative values force worst-case expected excess linear in $T$. We give PAC guarantees for learning compressed prediction tables, with efficient empirical-risk minimization for fixed landmarks. Our results connect metric coverage, prediction interfaces, and learning guarantees for compressed predictions in online MTS.

cs.LG↗

arXiv2026

From Trajectories to Grounded Preferences: Process Preference Synthesis via Interaction Element Graphs for Web PRMs

Comparative Process Reward Models (PRMs) provide critical step-level guidance for autonomous web agents by evaluating state-conditioned preferences between candidate actions. However, existing preference training data synthesized via multi-policy sampling suffers from a severe scarcity of Grounded Minimal Contrastive Pairs (GMCPs)-where competing candidates target genuine on-page elements with identical action types. In representative baselines preference data (namely, WebArbiter), GMCPs account for merely 24.19%, biasing PRMs during training to rely on shallow shortcuts (such as element hallucinations and action type mismatches) rather than acquiring genuine contextual decision semantics. To address these challenges, we propose SURFPRM, a graph-guided process preference synthesis framework for comparative Web PRMs. SURFPRM structures web demonstrations into a persistent Interaction Element Graph that acts as an environment-grounded negative action proposal mechanism, systematically synthesizing contrastive negative actions across spatial, temporal, and spatiotemporal confusion axes. This elevates the GMCP proportion from 24.19% to 74.60%, producing the curated SURFPRM-DATA dataset. Across six open-source backbones (3B to 9B parameters), PRMs trained on SURFPRM-DATA outperform baseline-trained models on average on WEBPRMBENCH and rival leading proprietary LLMs. In downstream reward-guided trajectory search on WEBARENA-LITE, SURFPRM provides step-level guidance for both GPT-4o (+14.21%) and GPT-4o-mini (+12.83%) policies, yielding substantial improvements in complex web task success rates.

cs.AI↗

arXiv2026

Separators in Enhancing Autoregressive Pretraining for Vision Mamba

The state space model Mamba has recently emerged as a promising paradigm in computer vision, attracting considerable attention for its efficient handling of long-sequence tasks. Its inherent causal structure makes it particularly well suited for autoregressive pretraining. However, existing autoregressive pretraining methods in vision are largely limited to short-sequence settings and may not fully use Mamba's capacity to model longer contexts. To investigate this setting, we introduce SeparaTors for AutoRegressive pretraining (STAR), a new autoregressive pretraining method for Vision Mamba that explicitly marks the boundaries between different images. STAR increases the patch-token sequence length from 144 to 640 by packing four images and four separator clusters. This is approximately $4.4\times$ the ARM patch-token sequence length. The increase is achieved without changing the resolution of any individual image: we use $192\times192$ inputs for autoregressive pretraining and $224\times224$ inputs for downstream classification fine-tuning. With this long-sequence pretraining scheme, STAR-B achieves 83.5\% EMA top-1 accuracy on ImageNet-1K after 1,600 epochs of pretraining. The learned representation also transfers beyond in-distribution classification: compared with ARM, STAR-B improves COCO box AP from 46.11 to 46.84 and mask AP from 40.74 to 41.45, while raising the mean top-1 accuracy across five ImageNet robustness benchmarks from 55.1\% to 56.8\%. Under the evaluated four-image setting, these results indicate that separator-based long-sequence pretraining improves recognition robustness and dense visual prediction relative to ARM.

cs.CV↗

arXiv2026

Approximating the Trace Distance Between Product Quantum States

We study the trace distance \[D_{\mathrm{tr}}(ρ,σ) =\frac12\|ρ-σ\|_1, ρ=\bigotimes_{i=1}^nρ_i,\quad σ=\bigotimes_{i=1}^nσ_i, \] when the two exponentially large states are specified by their local factors. We give a deterministic approximation within a universal constant factor for rational product inputs. Its running time is polynomial in the number of factors, the local dimension, and the input bit length. In the opposite direction, exact computation is $\#\mathsf P$-hard even for diagonal qubit states, by the corresponding hardness of total variation distance between product distributions. The proof uses local Uhlmann-optimal purifications to reduce the problem to estimating the product-fidelity defect and the trace norm of a structured first-order operator. Although this operator acts on an exponentially large space, we approximate its trace norm by a local convex surrogate that admits a polynomial-size classical conic formulation. A square-function estimate shows that the surrogate upper-bounds this trace norm. Conversely, duality and local dephasing reduce the reverse comparison to a head--tail inequality for independent centered random variables, showing that the surrogate is at most a dimension-free constant times the same norm.

cs.DS↗

arXiv2026

Self-Evolving Just-In-Time Memory for Proactive Embodied Safety

While Vision-Language Models (VLMs) have empowered embodied agents to execute complex household tasks, they struggle to proactively handle dynamically emerging hazards during closed-loop interactions. Existing safety approaches often rely on runtime guardrails to block unsafe actions or induce excessive caution, which severely stalls task progress instead of actively resolving the underlying risks. To break this safety-progress trade-off, we introduce the Self-Evolving Just-In-Time Memory framework, which reframes embodied safety from progress-stalling guardrails to proactive hazard mitigation. The framework consists of a Risk-Sufficient Topological Belief Graph (RSG) for persistent safety-relevant state tracking under partial observability, an Agency-Grounded Factual Memory for precise hazard anticipation, and an Experience Memory that injects procedural Meta-Skills to guide executable, progress-preserving mitigation. Furthermore, we propose an automated Test-Verify-Write loop, allowing agents to continually refine their mitigation Meta-Skills from execution traces at test time. Experiments on IS-Bench demonstrate that our framework substantially boosts the Safe-Success rate across multiple VLM backbones (e.g., +30.3% on Qwen3-VL-8B), enabling agents to proactively mitigate hazards without stalling task progress. Code is available at https://github.com/DyMessi/JIT-Memory.

cs.LG↗

arXiv2026

TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment

Reinforcement learning (RL) has shown extraordinary potential in aligning diffusion models to downstream tasks, yet most of them still suffer from significant reward hacking, which degrades generative diversity and quality by inducing visual mode collapse and amplifying unreliable rewards. We identify the root cause as the mode-seeking nature of these methods, which maximize expected reward without effectively constraining probability distribution over acceptable trajectories, causing concentration on a few high-reward paths. In contrast, we propose Trajectory Matching Policy Optimization (TMPO), which replaces scalar reward maximization with trajectory-level reward distribution matching. Specifically, TMPO introduces a Softmax Trajectory Balance (Softmax-TB) objective to match the policy probabilities of K trajectories to a reward-induced Boltzmann distribution. We prove that this objective inherits the mode-covering property of forward KL divergence, preserving coverage over all acceptable trajectories while optimizing reward. To further reduce multi-trajectory training time on large-scale flow-matching models, TMPO incorporates Dynamic Stochastic Tree Sampling, where trajectories share denoising prefixes and branch at dynamically scheduled steps, reducing redundant computation while improving training effectiveness. Extensive results across diverse alignment tasks such as human preference, compositional generation and text rendering show that TMPO improves generative diversity over state-of-the-art methods by 9.1%, and achieves competitive performance in all downstream and efficiency metrics, attaining the optimal trade-off between reward and diversity.

cs.LG↗

arXiv2026

Glove2Hand: Synthesizing Natural Hand-Object Interaction from Multi-Modal Sensing Gloves

Understanding hand-object interaction (HOI) is fundamental to computer vision, robotics, and AR/VR. However, conventional hand videos often lack essential physical information such as contact forces and motion signals, and are prone to frequent occlusions. To address the challenges, we present Glove2Hand, a framework that translates multi-modal sensing glove HOI videos into photorealistic bare hands, while faithfully preserving the underlying physical interaction dynamics. We introduce a novel 3D Gaussian hand model that ensures temporal rendering consistency. The rendered hand is seamlessly integrated into the scene using a diffusion-based hand restorer, which effectively handles complex hand-object interactions and non-rigid deformations. Leveraging Glove2Hand, we create HandSense, the first multi-modal HOI dataset featuring glove-to-hand videos with synchronized tactile and IMU signals. We demonstrate that HandSense significantly enhances downstream bare-hand applications, including video-based contact estimation and hand tracking under severe occlusion.

cs.CV↗

arXiv2026

Position: The Turing-Completeness of Autoregressive Transformers Relies Heavily on Context Management

Many works make the eye-catching claim that Transformers are Turing-complete. However, the literature often conflates two distinct settings: (i) a fixed Transformer system setting, in which a fixed autoregressive Transformer is coupled with a fixed context-management method to process inputs of different lengths step by step, and (ii) a scaling-family setting, in which a family of different models (with increasing context-window length or numerical precision) is used to handle different input lengths. Existing proofs of Transformer Turing-completeness are frequently established in setting (ii), whereas real-world LLM deployment and the standard notion of Turing-completeness correspond more naturally to setting (i). In this paper, we first formalize the fixed-system setting, thereby providing a concrete characterization of how real-world LLMs operate. We then argue that results proved in the scaling-family setting provide theoretically meaningful resource bounds but do not establish Turing-completeness, thereby clarifying a common misinterpretation of existing results. Finally, we show that different context-management methods can yield sharply different computational power, and we advocate the position that context management is a central component that critically determines the computational power of real-world autoregressive Transformers.

cs.AI↗