arXiv subjects

Kang Chen

Publications and source records attributed to Kang Chen.

At least 19 recordsLinked to original sources

arXiv2026

LaWAM: Latent World Action Models for Efficient Dynamics-Aware Robot Policies

Vision-Language-Action models (VLAs) leverage large-scale vision-language pretraining for semantic robot control, but often lack explicit foresight into how robot actions change the scene. World-Action Models (WAMs) address this limitation by conditioning policies on predicted futures, yet existing approaches typically rely on computationally expensive video generation with substantial pixel-level redundancy. We present LaWAM, a Latent World Action Model that exposes predictive dynamics to robot policies through compact latent visual subgoals instead of reconstructed future video. At the core of LaWAM is a latent-action-conditioned Latent World Model (LaWM). We obtain LaWM by training a latent action model in the latent space of a pretrained vision foundation model and repurposing its forward decoder to predict future observation features for scene evolution. LaWAM then conditions action generation on these predicted latent visual subgoals to enable dynamics-aware robot control. LaWAM achieves state-of-the-art or competitive success rates (SRs) across LIBERO (98.6% SR), RoboTwin (91.22% SR), and real-world manipulation tasks while retaining low-latency inference. LaWAM runs in 187 ms per action-chunk prediction and achieves up to 24x lower wall-clock latency than pixel-space WAMs.

cs.RO↗

arXiv2026

MetaDB: Putting File-System Structure Back into Distributed Databases

Storage systems in production increasingly rely on distributed databases for metadata management, yet our experience at Baidu shows that treating a general-purpose database as a black box leaves substantial performance and efficiency untapped. File-system metadata exhibits directory-tree locality, differentiated consistency requirements, and clustered deletions that general-purpose databases largely ignore. We thus propose MetaDB, a metadata-aware distributed transactional database that serves as the shared metadata layer for Baidu's large-scale storage systems. MetaDB addresses three key mismatches between general-purpose databases and metadata workloads: Data Model Mismatch, Transaction Model Mismatch, and Storage Implementation Mismatch. To bridge these gaps, MetaDB combines (1) tree-topology-induced sharding and execution, (2) semantic-aware off-critical-path processing, and (3) a lifecycle-aware storage engine with in-memory MVCC and locality-aware compaction. MetaDB achieves 2.2x-4.1x the throughput of our previous general-purpose database and halves the number of metadata machines. Deployed in production at Baidu for over five years, MetaDB unifies metadata management across the company's storage fleet and scales to over 1,000 metadata nodes, tens of trillions of metadata records, and exabyte-scale storage in a single cluster.

cs.DB↗

arXiv2026

Beyond Imitation: Reinforcement Learning-Based Sim-Real Co-Training for VLA Models

Simulation offers a scalable and low-cost way to enrich vision-language-action (VLA) training, reducing reliance on expensive real-robot demonstrations. However, most sim-real co-training methods rely on supervised fine-tuning (SFT), which treats simulation as a static source of demonstrations and does not exploit large-scale closed-loop interaction. Consequently, real-world gains and generalization are often limited. In this paper, we propose an RL-based sim-real Co-training (RL-Co) framework that leverages interactive simulation while preserving real-world capabilities. Our method follows a generic two-stage design: we first warm-start the policy with SFT on a mixture of real and simulated demonstrations, then fine-tune it with reinforcement learning in simulation while adding an auxiliary supervised loss on real-world data to anchor the policy and mitigate catastrophic forgetting. We evaluate our framework on four real-world tabletop manipulation tasks using two representative VLA architectures, OpenVLA and $π_{0.5}$, and observe consistent improvements over real-only fine-tuning and SFT-based co-training, including +24% real-world success on OpenVLA and +20% on $π_{0.5}$. Beyond higher success rates, RL co-training yields stronger generalization to unseen task variations and substantially improved real-world data efficiency, providing a practical and scalable pathway for leveraging simulation to enhance real-robot deployment.

cs.RO↗

arXiv2026

Strain Waves Generate Compression and Trigger Buckling in Active Polymers

Chemical reactions can sustain polymer motion without necessarily generating the compression required for buckling. Theory and simulations show that tension-dependent chemical transitions coupled to bond elasticity generate finite-wavelength traveling strain waves in a semiflexible polymer. In the straight chain at fixed axial span, the waves reduce mean active contraction while leaving mean elastic tension unchanged, generating compression that drives bending above a threshold. Releasing the mean load allows longitudinal waves to persist while transverse perturbations decay.

cond-mat.soft↗

arXiv2026

Complex pattern in periodically driven binary Quincke colloids

Binary active systems offer a minimal platform for understanding how component heterogeneity organizes non-equilibrium structures. Here we show that in a binary Quincke system driven by a periodic square-wave electric field, a dynamic square lattice in which small particles occupy the lattice-cell centers which are surrounded by large particles emerges spontaneously. Particle-resolved trajectories reveal that the size separation arises from the mobility difference of large particles and small particles. Increasing the driving frequency reduces the single-cycle path lengths of both species and simultaneously decreases the lattice constant. These results show how size-dependent activity,density redistribution, and periodic forcing cooperate to generate tunable complex active patterns.

cond-mat.soft↗

arXiv2026

Optimal Recursive Composition and Dyadic Phase Laws for Gradient Descent with Predetermined Stepsizes

Predetermined stepsize schedules featuring carefully chosen long steps have recently been shown to accelerate gradient descent (GD) on smooth convex functions. A prominent class of such schedules is built through recursive composition. In this paper, we characterize the convergence of these optimized recursive schedules, revealing a nonconstant log-periodic modulation across prescribed horizons. Specifically, we prove that for every $N \geq 1$, the optimized ConPP/OBS-S schedules satisfy $$f(x_{N-1})-f_\ast \le \frac{1}{2N^pΦ(\log_2N)-1} \frac{L}{2}\|x_0-x_\ast\|^2, \qquad p=\log_2(1+\sqrt2), $$ where $Φ$ is a positive, Lipschitz, nonconstant $1$-periodic function. We derive this by proving that balanced splitting is optimal at every horizon for these constructions, resolving a conjecture of Zhang and Jiang. Furthermore, for their asymmetric counterparts (ConPD/OBS-F schedules), we show that although optimal splits are not necessarily balanced, the same Silver exponent asymptotically persists alongside a distinct log-periodic modulation.

math.OC↗

arXiv2026

STEAM: Self-Supervised Temporal Ensemble Advantage Modeling for Real-World Robot Learning

Real-world robot learning increasingly relies on heterogeneous data, but demonstrations and rollouts often mix useful progress with stalls, corrections, and suboptimal behavior. Effective policy learning therefore requires frame-level advantages that distinguish reliable local progress from failures and regressions. We propose \textbf{S}elf-supervised \textbf{T}emporal \textbf{E}nsemble \textbf{A}dvantage \textbf{M}odeling (\textbf{STEAM}), a label-free method that learns such advantages from expert demonstrations. STEAM trains an ensemble of temporal-offset predictors on frame pairs within expert trajectories, using the normalized temporal offset between two frames as a self-supervised signal. Each predictor maps a frame pair to a distribution over temporal offsets, which is converted into a scalar advantage. STEAM then takes the minimum advantage across the ensemble to score mixed-quality rollout data conservatively. Across real-world bimanual towel folding, chip checkout, table clearing, and cola restocking tasks, STEAM identifies stalls, failures, and recoveries. When combined with CFGRL, STEAM achieves the highest policy success rate on all four tasks, averaging $83.5\%$ against $58.8\%$ for the best baseline. Project page: https://rlinf.github.io/steam/.

cs.RO↗

arXiv2026

Spontaneous Motion Generates Reversible Nonreciprocity in an Achiral Active Elastic Ring

Nonreciprocal mechanical response is usually associated with built-in directional couplings, structural chirality, or external driving. Whether an achiral active body can instead generate and reverse such directionality through its own motion remains less clear. Here we show that spontaneous rotation makes an active elastic ring with reciprocal passive interactions mechanically nonreciprocal. Off-center elastic forces reorient propulsion, coupling deformation back to motion. Reversing the rotation reverses the antisymmetric response, while a compensating control preserves finite directionality but strongly suppresses the near-Hopf resonance. Approaching Hopf, the selected response grows inversely with distance to onset and becomes a self-sustained traveling deformation above threshold. Thus nonreciprocity is selected by the dynamical state, while active feedback controls its critical amplification.

cond-mat.soft↗

arXiv2026

When Keywords Drop but Classifiers Hold: Soft Refusals under KV Cache Compression

KV cache compression is widely used for long context LLM inference under memory constraints, while deployed systems typically score refusals after generation with keyword filters or learned classifiers. Such monitors are intended to indicate whether a model declined a harmful request under the serving regime actually used. However, it remains unclear whether matched compression that preserves task accuracy also preserves agreement between lightweight lexical monitors and stronger refusal classifiers. We study this with a paired protocol on n=200 harmful prompts with a long filler context: each prompt is answered once under full retention and once under matched eviction after a shared prefill, and the same replies are scored by keyword heuristics, the HarmBench Llama-2-13B classifier, an auxiliary LLM judge, and humans on disagreements. On Qwen2.5-3B, keyword refusal falls from 98.0% to 80.5% (McNemar p~1e-8) while classifier refusal stays near ceiling (99.0%-99.5%) and MMLU accuracy is unchanged (50.0%); human labels predominantly follow the classifier, consistent with soft refusals. The gap is not universal and weakens under short fillers and paired SnapKV, so safety auditing under compression should rely on several judges matched to the serving context rather than on keyword rates alone.

cs.LG↗

arXiv2026

Data Security in Large Language Models: Risks, Defense, and Directions

Large Language Models (LLMs), now a foundation in advancing natural language processing, power applications such as text generation, machine translation, and conversational systems. Despite their transformative potential, these models inherently rely on massive amounts of training data, often collected from diverse and uncurated sources, which exposes them to serious data security risks. Harmful or malicious data can compromise model behavior, leading to toxic outputs or hallucinations, while also creating vulnerabilities to data-driven attacks such as prompt injection and data poisoning. As LLMs continue to be integrated into critical real-world systems, understanding and addressing these data-centric security risks is imperative to safeguard user trust and system reliability. This survey offers a comprehensive overview of the main data security risks facing LLMs and reviews current defense strategies, including adversarial training, data cleaning, output guardrails, Reinforcement Learning from Human Feedback (RLHF), data augmentation, and Retrieval-Augmented Generation (RAG)/agent defenses. Additionally, we categorize and analyze relevant datasets used for assessing robustness and security across different domains, providing guidance for future research. Finally, we highlight key research directions that focus on data provenance and traceability, verifiable machine forgetting, secure model updates, standardized evaluation framework, explainability-driven security analysis, and effective governance frameworks, aiming to promote the safe and responsible development of LLM technology. This work seeks to inform researchers, practitioners, and policymakers, driving progress toward data security in LLMs.

cs.CR↗

arXiv2026

The Exact Time-Uniform Rate Frontier for Stochastic Gradient Descent on Smooth Convex Objectives

We study the time-uniform convergence of the raw iterate of standard stochastic gradient descent (SGD) for unconstrained smooth convex objectives. We prove that, under standard noise assumptions, the time-uniform convergence rate gets arbitrarily close to $\sqrt{\log n / n}$ but never reaches it. More specifically, we prove that for every positive, eventually nondecreasing sequence $h$ satisfying $h(n) = o(\sqrt{n})$, a bound of order $h(n)/\sqrt{n}$, holding simultaneously for all $n$ with probability at least $1-α$ and uniformly over the problem class, is achievable if and only if \[ \sum_{j = 1}^{\infty} \frac{1}{h(2^j)^2} < \infty. \] The constructive sufficiency result follows from a dyadic horizon-free schedule together with an additive conditional-restart inequality. The necessity counterpart applies to every deterministic nonnegative schedule and holds even for a one-dimensional analytic smooth convex objective with Gaussian noise.

math.OC↗

arXiv2026

From Concentration to Differentiation and Back: Routing Effective Rank in MoE Reasoning Cohorts

Test-time scaling produces cohorts of reasoning rollouts, yet there is no standard label-free account of how their internal computation reorganizes as inference unfolds. We introduce routing effective rank deff, the entropy-effective dimensionality of a cross-rollout graph built from MoE expert-routing similarity. Across ten MoE configurations and five math/science benchmarks, deff exhibits a reproducible low-high-low trajectory, with a prominent interior maximum in 98.5% of 3,105 model-question cohorts: routing similarity is concentrated early, maximally differentiated at intermediate budgets, and reconcentrated later, and the timing of this maximum varies systematically with architecture and reasoning effort. An exact decomposition separates cohort-wide common-mode mass from residual spectral dimensionality: common-mode reallocation accounts for about two thirds of the trajectory, while the residual spectrum contributes about one quarter and retains substantial variation beyond the common mode. The decomposition further localizes behavior: among non-unanimous cohorts, increases in common-mode concentration strongly predict same-answer recoverability, and higher reasoning effort delays the maximum by 2.59 octaves (doublings of the token budget) and consistently expands the high-rank period across all four tested architectures, locating the effort effect in timing and duration rather than peak amplitude. Correctness comparisons separate structural monitoring from answer selection, positioning routing effective rank as a decomposable, label-free diagnostic of cohort organization - a principled spectral lens on how MoE reasoning cohorts differentiate and reconcentrate over inference time.

cs.AI↗

arXiv2026

Beyond the Trace: Coupling an Interpretable Reasoning-State Readout to Native MoE Routing

What a reasoning model writes is only a partial record of the process that produces it. We introduce a two-level internal readout for mixture-of-experts reasoning. We first distill vocabulary-scale J-space into J64, a 64-axis semantic frame learned from the model's own reasoning states. J64 reveals readable process state that the emitted trace does not show: it separates inference effort from problem-induced strain. It also adds 0.096 to 0.135 held-out AUC over a baseline that reads the same rollout as token occupancy and aggregates it in exactly the same way. We then reconstruct J64 from native expert-routing statistics. The result is R64, a low-overhead proxy: its median per-axis correlation with J64 is 0.69 to 0.86 across three models and two families, and on gpt-oss-20b it preserves 95 to 100% of J64's predictive gain. The readout supports test-time decisions at two temporal resolutions. Over completed candidate sets, J64 and R64 improve single-branch selection, and R64-weighted voting improves plain majority voting in seven of eight settings. During generation, rolling readout windows drive a cumulative stop-and-resample policy whose operating point is fixed on training questions alone. J64 improves accuracy by 1.1 to 5.9 points over a sibling-permuted control, and the routing-only R64 proxy retains 0.9 to 3.2 of those points. Finally, router edits aimed at the mechanism J64 names induce the predicted reasoning behaviors and shift a diagnosed stall from numerical guessing toward exact symbolic execution. Together, J64 makes latent process state readable, while routing makes it deployable and actionable.

cs.AI↗

arXiv2026

Disagree to Explore, Agree to Commit: Routing-Guided Test-Time Scaling for Software Agents

Software-engineering agents solve repository-level tasks through long, stochastic tool-use trajectories, and repeated attempts often find fixes missed by one run. Test-time scaling is difficult because patches lack canonical answer forms, while sibling actions from a shared prefix are correlated. We study whether native MoE router traces can guide steering and selection without an external judge or selection-time test execution. Our analysis shows that routing provides a robust behavioral role signal; token-granular readouts and decision-matched comparison sets turn it into effective control. We therefore introduce Risa (Routing-Informed Steering and Arbitration): within trajectories, routing encourages diverse exploration and controlled convergence during patch commitment; across separately sampled trajectories, agreement at informative patch positions selects a final candidate. We evaluate on SWE-bench Verified using open-weight sparse MoE agents across scales and reasoning-effort settings. Risa's routing arbitration raises the macro-average resolved rate from 44.9% under uniform sampling to 48.2% on the gpt-oss family, matching text consensus without answer-string matching, and it transfers to Qwen3.6, where it improves on uniform choice and matches text consensus on the full 500-task benchmark.

cs.AI↗

arXiv2026

FORCE-Interior: Measurement-Consistent Adaptation of a Poisson-Flow Generative Prior for Interior CT

Interior tomography reconstructs a region of interest (ROI) from truncated projections, an ill-posed problem with non-unique solutions and truncation-induced bias. Existing deep-learning methods can be sensitive to changes in ROI geometry and noise, while expressive generative priors may produce measurement-inconsistent content without measurement constraints. We propose FORCE-Interior, a training-free adaptation of a pretrained Poisson-flow generative prior to interior CT. A full-field-of-view (FOV) OS-SART warm start avoids forcing all measured attenuation into the ROI, and truncation-mask-aware OS-SART updates enforce data consistency throughout sampling. In our experiment, FORCE-Interior achieves the best PSNR, SSIM, and LPIPS at the two more severely truncated synthetic ROI sizes and competitive performance at the largest ROI, while maintaining low projection-domain residuals. These findings support the measurement-consistent adaptation of a reusable generative CT prior, while further clinical and patient-level validation remains necessary.

eess.IV↗

arXiv2026

RLinf-VLA: A Unified and Efficient Framework for Reinforcement Learning of Vision-Language-Action Models

Recent studies have demonstrated the potential of reinforcement learning (RL) to improve the task performance of vision-language-action (VLA) models through interaction. However, current efforts remain fragmented, lacking a unified platform for fair comparison across architectures and algorithms, as well as an efficient system design for scalable training. Therefore, we present RLinf-VLA, a unified and efficient framework for scalable RL training of VLA models. RLinf-VLA standardizes the integration of diverse VLA architectures, RL algorithms, and heterogeneous simulators through a unified interface, enabling extensibility and reproducibility. To improve efficiency, the framework adopts a flexible resource allocation architecture for rendering, inference, and training in RL pipelines. In particular, RLinf-VLA introduces a hybrid fine-grained pipeline allocation strategy that achieves a 1.61$\times$-1.88$\times$ training speedup on ManiSkill. Using this framework, RL-trained models achieve strong performance across embodied benchmarks, including 98.11% success on 130 LIBERO tasks, 97.66% success on 25 ManiSkill tasks, and 84.63% average success across 6 RoboTwin tasks. In addition, RLinf-VLA distills a set of effective practices for RL-based VLA training. We envision RLinf-VLA as a foundational framework for efficient, unified, and reproducible research in embodied intelligence.

cs.RO↗

arXiv2026

Ge$^\text{2}$mS-T: Multi-Dimensional Grouping for Ultra-High Energy Efficiency in Spiking Transformer

Spiking Neural Networks (SNNs) offer superior energy efficiency over Artificial Neural Networks (ANNs). However, they encounter significant deficiencies in training and inference metrics when applied to Spiking Vision Transformers (S-ViTs). Existing paradigms including ANN-SNN Conversion and Spatial-Temporal Backpropagation (STBP) suffer from inherent limitations, precluding concurrent optimization of memory, accuracy and energy consumption. To address these issues, we propose Ge$^\text{2}$mS-T, a novel architecture implementing grouped computation across temporal, spatial and network structure dimensions. Specifically, we introduce the Grouped-Exponential-Coding-based IF (ExpG-IF) model, enabling lossless conversion with constant training overhead and precise regulation for spike patterns. Additionally, we develop Group-wise Spiking Self-Attention (GW-SSA) to reduce computational complexity via multi-scale token grouping and multiplication-free operations within a hybrid attention-convolution framework. Experiments confirm that our method can achieve superior performance with ultra-high energy efficiency on challenging benchmarks. To our best knowledge, this is the first work to systematically establish multi-dimensional grouped computation for resolving the triad of memory overhead, learning capability and energy budget in S-ViTs. Code is available at https://github.com/hzc1208/Ge2mST.

cs.NE↗

arXiv2026

Seen, Said, or Forgotten? A Causal Audit of Visual KV Memory Across Dialog Turns

Stateful multimodal assistants encode an image once but may answer questions about it many turns later. Attention-guided visual-KV eviction assumes that evidence irrelevant now will remain dispensable, although future questions are unknown. We ask when a visual fact is actually safe to forget and introduce the Causal Visual Memory Audit (CVMA), a paired single-prefill framework that tests what later answers lose when a visual region, the whole image, or prior assistant text becomes unavailable. On VisDial and ConvBench, current attention can rank future-useful regions worse than random even though a diagnostic marginal-utility control shows substantial selection headroom. Aggregate scores hide this failure when later turns do not need vision; controlled and stock-generated histories reveal a second escape route, in which assistant-text KV replaces image KV for facts already stated but not reliably for unstated facts. In the tested stacks, safe forgetting is supported by low future visual dependence or fact-specific verbalization---not by low current attention.

cs.CV↗