arXiv subjects

Kai Wang

Publications and source records attributed to Kai Wang.

At least 73 records · Page 4Linked to original sources

arXiv2026

HEAL: Hindsight Entropy-Assisted Learning for Reasoning Distillation

Distilling reasoning capabilities from Large Reasoning Models (LRMs) into smaller models is typically constrained by the limitations of rejection sampling. Standard methods treat the teacher as a static filter, discarding complex "corner-case" problems where the teacher fails to explore valid solutions independently, thereby creating an artificial "Teacher Ceiling" for the student. In this work, we propose Hindsight Entropy-Assisted Learning (HEAL), an RL-free framework designed to bridge this reasoning gap. Drawing on the educational theory of the Zone of Proximal Development (ZPD), HEAL synergizes three core modules: (1) Guided Entropy-Assisted Repair (GEAR), an active intervention mechanism that detects critical reasoning breakpoints via entropy dynamics and injects targeted hindsight hints to repair broken trajectories; (2) Perplexity-Uncertainty Ratio Estimator (PURE), a ratio-based filtering heuristic that reduces high-anomaly shortcut-like rationales; and (3) Progressive Answer-guided Curriculum Evolution (PACE), a three-stage distillation strategy that organizes training from foundational alignment to hard-case adaptation. Extensive experiments on multiple benchmarks demonstrate that HEAL significantly outperforms traditional SFT distillation and other baselines.

cs.AI↗

arXiv2026

EvoSQL: Memory-Augmented Critic-Generator Co-Evolution for Text-to-SQL

Text-to-SQL has advanced rapidly with large language models, but complex database queries still require reasoning beyond one-shot generation, including multi-step decomposition, execution-based diagnosis, and targeted correction. We present EvoSQL, a co-evolution framework that formulates SQL synthesis as an iterative interaction between a generator and a critic. EvoSQL maintains a contextualized candidate memory, verifies SQL candidates with both execution signals and LLM-based critique, and updates its memory through utility-guided aggregation. To strengthen the underlying generator-critic pair, we further introduce a Self-Distillation Policy Optimization (SDPO) fine-tuning stage that injects execution-aware supervision into modern coding LLM backbones. Experiments on Spider and BIRD show that EvoSQL consistently improves open-source models over Maj@16 baselines, with particularly large gains on BIRD-Dev, ranging from +1.37% for Qwen3-4B to +9.19% for Qwen2.5-Coder-3B. SDPO initialization further improves selected backbones on Spider-Test and BIRD-Dev. These results suggest that memory-grounded co-evolution is an effective path toward more reliable and generalizable Text-to-SQL systems. Code is available at https://github.com/valleysprings/EvoSQL.

cs.AI↗

arXiv2026

Is Multimodal Speculative Decoding Ready for Diffusion-Based Parallel Drafting? A Survey and Empirical Diagnosis

Speculative decoding accelerates autoregressive generation by allowing a lightweight drafter to propose future tokens while a target model verifies them in parallel. Its lossless guarantee has motivated a line of work that pushes the drafter itself toward parallel generation. The most recent paradigm is block-parallel generative drafting, including diffusion-based methods such as DFlash and DSpark, achieving up to 3.6x speedup on common daily chatting tasks. While this transition is well studied in text-only LLMs, its applicability to multimodal models remains an open question. Existing multimodal speculative decoding efforts focus on input compression, adapter alignment, candidate coverage, or modality-specific verification; however, block-parallel generative drafting remains largely unexplored. To bridge this gap, this paper combines a modality-centered survey with a cross-architecture empirical study to ask: Is multimodal speculative decoding ready for diffusion-based parallel drafting? In this survey, we systematically analyze a wide spectrum of multimodal models, spanning Vision-Language, Video-Language, Audio, and Vision-Language-Action (VLA) architectures, from the dual perspectives of drafting parallelism and cross-modal information interaction. We introduce a unified taxonomy that isolates drafter-side parallelism from orthogonal design choices such as tree construction and verification strategies. Furthermore, we provide a comprehensive empirical comparison of existing methods under varying degrees of parallelism across standardized multimodal benchmarks, including OCR, VQA, visual reasoning, and image captioning. Finally, we summarize the limitations of current approaches, discuss open challenges, and outline promising future directions for this rapidly evolving field.

cs.AI↗

arXiv2026

GhostTac: Manipulating Tactile Sensors without Physical Contact

Tactile sensors are integral components of modern robotic systems, enabling robots to perceive and interact with the physical environment through tactile feedback. Despite their importance, the physical-layer security of tactile sensors has received little attention in prior work. In this paper, we present GhostTac, to the best of our knowledge, the first contactless attack that manipulates tactile sensing via electromagnetic interference (EMI). We identify that EMI exploits the nonlinear rectification and limited bandwidth amplification effects, allowing carefully crafted EMI signals to be converted into a persistent DC offset that bypasses on-board filtering and induces stable measurement deviations. Building on this mechanism, GhostTac enables fine-grained and controllable manipulation of sensor outputs by reshaping the spatial distribution and manipulating the magnitude at the targeted location. Such interference can induce unintended and harmful robot behaviors, such as causing a domestic robot to exert excessive force, resulting in physical damage or human injury. We evaluate GhostTac on 10 sensor modules and 2 dexterous hands, covering 15 tactile sensors of different types, and demonstrate consistent attack effectiveness across all tested devices. We further present three case studies on tactile grasping, slip detection, and material classification to illustrate practical impacts in real robotic tasks. We envision that our findings shed light on a new physical attack vector against tactile sensing in robotic systems.

cs.CR↗

arXiv2026

SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces

Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files, tools, memory, and execution environments. However, this modularity introduces attack surfaces that are largely missed by existing safety evaluations: even when the user request is benign, unsafe influence may reside in skill guidance, local artifacts, or execution-environment files that steer the agent toward unsafe actions. We present SkillSafetyBench, a runnable benchmark for evaluating such skill-facing safety failures. SkillSafetyBench includes 155 adversarial cases across 47 tasks, 6 risk domains, and 30 safety categories, each evaluated with a case-specific rule-based verifier. Experiments with multiple CLI agents and model backends show that non-user attacks can consistently induce unsafe behavior, with distinct failure patterns across domains, attack methods, and scaffold-model pairings. Our findings suggest that agent safety depends not only on model-level alignment, but also on how agents interpret skills, trust workflow context, and act through executable environments. The complete benchmark is available at https://github.com/AI45Lab/skill-safety-bench.

cs.CR↗

arXiv2026

Scalable Question-Centric Text-to-Image Evaluation: Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing

Modern text-to-image (T2I) models often have similar total scores but different strengths, making practical selection difficult. Fine-grained benchmarks decompose prompts into questions, yet often return them to prompt scores and fixed categories, weakening attribution and ignoring complexity. Related requirements are also scored separately or as one total, obscuring basic versus compositional failure. We present QC-T2I-Bench, a question-centric framework that converts open prompts into attributed atomic questions and organizes their dependencies with Davidsonian Scene Graphs (DSGs). We use hierarchy-constrained question aggregation to exclude downstream questions after a prerequisite fails and to prevent simple and complex prompts from receiving the same total weight. We then use the DSG structure to measure joint success within prompts and compare repeated entities across prompts, separating basic realization failures from failures under additional requirements. We evaluate multiple open-source T2I models on English and Chinese prompts. The resulting question-level evidence supports reliable ranking and fine-grained diagnosis: joint completion falls from 80.7\% for components with two capabilities to 37.2\% for those with seven or more. Finally, we reuse the same records for training-free routing; our cost-aware router matches ERNIE's 89.51-point estimate with 21.3\% less GPU-s/MP.

cs.AI↗

arXiv2026

Read Less, Solve More: Token-Efficient Sparse Reading for AI Agents

Long-horizon agents increasingly rely on repeated access to external artifacts, yet current reading interfaces often expose entire objects even when only sparse evidence is needed. This over-reading increases token and latency costs and can dilute task-relevant evidence, while existing context-reduction methods mainly intervene after broad content has already entered the trajectory. We present SparseRead, a training-free, model-transparent reading layer that controls content admission before unnecessary evidence reaches the model context. SparseRead combines a regime-aware Read Gate, extensible Reader Backends, and a stateful protocol for bounded, source-anchored evidence acquisition with explicit refinement, verification, stopping, and fallback. Across six frontier models, including Claude Opus 5, and five workload scenarios, SparseRead reduces token volume by up to 92.9% and wall time by up to 89.0%, while preserving or improving task quality. Its consistent gains across three agent frameworks further demonstrate broad portability.

cs.AI↗

arXiv2026

ClawSentry: A Progressive Multi-Tier Security Monitor for Safeguarding Autonomous LLM Agents

As large language model (LLM) agents move from conversation to executing code, reading local files, and orchestrating external tools, a single agent hijacked by a malicious third-party skill can cause data exfiltration, privilege escalation, or cascading compromise. We argue that agentic risk is progressive: it can enter at four loci of the agent control loop--skill admission, invocation-time intent, execution-time effect, and post-action consequence--while a denied dangerous objective can reappear across surface forms, tools, or turns; existing safeguards are typically local to one lifecycle boundary or one call. Guided by this threat model, we present ClawSentry, an open-source, framework-agnostic security supervision gateway for agent runtimes. Before a skill package is ever executed, First-use Skill Package Review (FSPR) audits it under a deterministic evidence floor, escalating unresolved cases to bounded read-only agentic review (locus A). At runtime, a three-tier progressive decision engine--a deterministic L1 layer, a rule-anchored L2 semantic reviewer, and a read-only L3 evidence-seeking agent--spends contextual review only on the residual ambiguity, while a session-level anti-bypass mechanism recognizes tool-switching and rephrased retries (loci B--C); a post-action path feeds high-severity evidence non-retroactively into later review (locus D). An Agent Harness Protocol (AHP) abstraction applies one policy across Codex, Claude Code, Kimi CLI, and Gemini CLI without modifying agent internals. On SkillInject with Codex/GPT-5.4, contextual ASR falls from 39.55% to 2.61% while contextual TSR moves only from 83.78% to 83.05%. Across five Work Agents on the full SkillsSafety benchmark, ClawSentry confines ASR to 9.09--15.03% from 33.5--49.7% unprotected, and aggregate TSR on clean skills remains 98.7%.

cs.CR↗

arXiv2026

Revealed Default Under Choice Overload

Empirical and experimental studies suggest that a decision maker can remain with their default option even when preferred alternatives are available, yet standard choice data may not record which option is the default. Existing default-based models often assume that the default option is observed. In contrast, I study a decision maker subject to status quo bias and choice overload and illustrate how one can identify both the decision maker's default and preference from the observed behavior. This paper shows that anomalous choice behavior may arise from a latent default, a possibility largely overlooked in studies of standard choice data.

econ.TH↗

arXiv2026

A Symmetric Counterexample to the Snashall--Solberg Conjecture

It is shown that the trivial extension of the Xu--Snashall algebra is a symmetric counterexample to the Snashall--Solberg conjecture, which states that the Hochschild cohomology ring of a finite-dimensional algebra modulo nilpotence is a finitely generated algebra. To the best of our knowledge, this is the first known selfinjective counterexample.

math.KT↗

arXiv2026

Deformations and homotopy theory of Rota-Baxter Lie algebras

For Rota-Baxter Lie algebras, a homotopy cooperad is exhibited, whose cobar construction is shown to be the minimal model of the operad of Rota-Baxter Lie algebras by using algebraic Morse theory. The deformation complex of Rota-Baxter Lie algebras as well as the $L_\infty$-algebra structure on this complex are deduced from the minimal model and the notion of homotopy Rota-Baxter Lie algebras is given as a consequence.

math.KT↗

arXiv2026

Topologically Configurable Nonlinear Vortex Generation at van der Waals Heterostructures

van der Waals (vdW) materials offer a highly tunable and efficient platform at nanoscale for nonlinear and quantum optics. Twist-stacked vdW heterostructures enable elegant control of symmetry and interlayer coupling. Prior studies mainly focus on planar twisted interfaces, while neglecting the naturally formed and mandatory defects in such vdW heterostructures. Here, we demonstrate nonlinear singular optics with topologically configurable nonlinear vortex generation at the corner singularity of vdW heterostructures. By tailoring azimuthally discrete second-harmonic phase gradients at each interface, we obtain programmable nonlinear vortex emitters with dominant target OAM components. Nonlinear OAM beams with topological charge $\ell = 1$ and $\ell = -2$ are experimentally realized, respectively. Our work unlocks the untapped potentials of nonlinear singular optics in twisted vdW materials as a reconfigurable and lithography-free platform for nonlinear structured light generation, important in quantum nonlinear optics and related fields.

physics.optics↗

arXiv2026

StateM: Reaching 95.3% Raw Accuracy, or a \$15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling

Long-horizon agents can fail even when their underlying models can solve the constituent steps. They may lose track of mutable state, fail to reactivate lessons from earlier executions, skip known procedures, or stop prematurely. We bet on harness scaling to improve the execution system around an agent without changing its model weights. We introduce StateM, an agent-native runtime that organizes execution around durable states, phase-local context, checked transitions, recoverable runbooks, and versioned procedural practices that agents and users can inspect together. On Terminal-Bench 2.1, StateM raises GPT-5.5 xhigh to 92.1\%, versus 83.1\% reference and GPT-5.6 Sol Ultra at 91.9\%. The runbook transfers unchanged to GPT-5.6. With GPT-5.6 Sol xhigh, StateM reaches 95.3\% raw accuracy across 445 trials and succeeds on all 89 tasks at least once. The frozen profile raises GPT-5.6 Luna from 76.7 to 85.4\%, above the 84.9\% Sol xhigh reference. Using the same runtime, runbook structure, and golden rules, less than \$38 of adaptation raises DeepSeek-V4 Flash from 82.7 to 88.1\% under standard timeouts and to 89.1\% on an 88-task common core. Extending only the remaining latency-sensitive task matches the reported 88.8\% GPT-5.6 Sol max result. Final-score API usage is about \$15 versus \$574.68 for the GPT reference; total DeepSeek expenditure is \$52.22. On BusinessBench, family-specific runbooks built on development sets yield held-out gains of 0.55 macro and 1.34 micro points; two mechanism-matched families improve by 10.04 points. Concrete rules generalize when tasks share execution structure, while the control methodology applies broadly. StateM turns selected postmortem findings into persistent, executable preconditions and practices, making learned controls explicit and enforceable through stateful controls. Code at github.com/henryqin1997/statem.

cs.AI↗

arXiv2026

Practical Lossless Volumetric Medical Image Compression via Tri-plane Context Tree Learning

Lossless compression of volumetric medical images is of paramount importance for clinical and research applications where data fidelity is essential. Traditional compression methods are often limited in efficiency due to rigid, handcrafted models. Conversely, deep neural network (DNN)-based compression methods, while effective, demand substantial computational resources, hindering deployment in resource-constrained settings. To address these challenges, we propose a novel tri-plane context tree (TCT)-based method for lossless volumetric medical image compression that delivers high performance without relying on DNNs or external training data. To exploit intra-slice and inter-slice redundancies, we introduce a compact tri-plane context representation that decomposes complex 3D context modeling into efficient 2D modeling on three orthogonal planes. By integrating this representation with a context tree framework, we develop an input-specific TCT model employing an adaptive binary tree structure. At each tree node, the model dynamically selects from a suite of tri-plane based predictors and contextual feature extractors, enabling data-adaptive context modeling tailored to local structural characteristics. Instead of offline training, we sample a subset of the input volume to learn the TCT model by optimizing the minimum description length (MDL) through iterative construction and pruning. With the learned TCT model, each pixel retrieves its corresponding context, computes the prediction residual using the predictor dictated by the context, and performs entropy encoding based on the associated histograms. Experimental results demonstrate that the proposed method achieves compression performance on par with recent DNN-based methods on multiple datasets, while maintaining low computational cost and fast coding speeds, making it highly applicable in practice.

eess.IV↗

arXiv2026

The origin of the galaxy size-stellar metallicity relation: A semi-analytical perspective

Stellar metallicity encodes the integrated effects of gas inflow, star formation, and feedback-driven outflow, yet its connection to galaxy structure remains poorly understood. Using SDSS-IV MaNGA, we present the direct observational evidence that, at fixed stellar mass, smaller central galaxies are systematically more metal-rich, with a Spearman's rank correlation coefficient reaching $R_{\rm s}\approx -0.4$. The semi-analytical model L-GALAXIES reproduces this anti-correlation, albeit with a stronger amplitude ($R_{\rm s}\approx -0.8$). Within this framework, the trend cannot be explained by differences in gravitational potential depth or star formation history. Instead, smaller galaxies attain higher stellar metallicities because their elevated star formation efficiencies accelerate chemical enrichment, and, at fixed stellar mass, they inhabit less massive haloes, which makes their recycled inflows more metal-rich. The gas-regulator model demonstrates that star formation efficiency affects stellar metallicity when the system has not long remained in equilibrium, which is shown to be the case for central galaxies with $M_{\rm star}\lesssim 10^{10.5}\rm M_\odot$ in both L-GALAXIES and observation. The model also suggests a testable signature that, at fixed stellar mass, larger or lower-metallicity galaxies should inhabit more massive haloes than their smaller and higher-metallicity counterparts, providing a direct and testable imprint of the galaxy size-stellar metallicity relation on the galaxy-halo connection.

astro-ph.GA↗

arXiv2026

Gravitational potential drives the concentration dependence of the stellar mass-halo mass relation

We investigate the origin of the scatter in the stellar mass-halo mass (SMHM) relation using the \colibre cosmological hydrodynamical simulations. At fixed halo mass, we find a clear positive correlation between stellar mass and halo concentration, particularly in low-mass haloes between $10^{11}$ and $10^{12}\,\rm M_\odot$, where all halo properties are computed from the corresponding dark-matter-only simulation. Two scenarios have been proposed to explain this trend: the earlier formation of higher-concentration haloes allows more time for star formation, or the deeper gravitational potential wells of higher-concentration haloes enhance baryon retention. To distinguish between them, we examine correlations between halo concentration, stellar mass, stellar age, and stellar metallicity. While, at fixed halo mass, halo concentration correlates with stellar age, stellar age itself shows only a weak correlation with stellar mass, indicating that early formation alone cannot account for the concentration-dependence in the scatter of the SMHM relation. In contrast, both stellar metallicity and halo concentration exhibit correlations with stellar mass. The connection between halo concentration and stellar metallicity persists even when simultaneously controlling for both halo mass and stellar mass. These results support the scenario in which the deeper gravitational potentials in higher-concentration haloes suppress feedback-driven outflows, thereby enhancing both baryon and metal retention.

astro-ph.GA↗

arXiv2026

Experimental quantum telecloning across silicon photonic chips

Telecloning -- the combination of quantum teleportation and cloning -- offers a powerful mechanism to disseminate unknown quantum states to multiple spatially separated recipients with optimal fidelity. Despite its conceptual importance for quantum networks, an experimental demonstration of symmetric qubit quantum telecloning remains elusive, particularly due to the challenges of generating multipartite entangled resource states and implementing stable multi-photon interference across distributed nodes. Here, we realize the optimal 1 to 2 symmetric quantum telecloning using a scalable silicon photonic platform. We implement a six-photon protocol using two independent, fiber-linked photonic chips: one generating a heralded input state and the other preparing a four-photon entangled resource state. By performing an interchip Bell-state measurement, we successfully distribute the input state into two optimal clones at remote nodes. We observe an interchip cloning fidelity of 78.45 $\pm$ 1.39%, exceeding the classical limit of 2/3 by 8 standard deviations. Our results demonstrate the robust generation and manipulation of complex multi-photon states between integrated chips, providing a foundational building block for large-scale multi-party quantum networks.

quant-ph↗

arXiv2026

Photonic realization of a subgraph extraction in a quantum random network

Understanding how complex connectivity emerges in networks is a fundamental challenge in classical and quantum science. In classical random networks, complex subgraphs typically require relatively high connection probabilities, whereas quantum random network theory predicts that such structures can arise at a single, lower threshold through entanglement and local operations. Here, using an integrated silicon photonic chip, we experimentally realize a quantum subgraph predicted by quantum random network theory in a four-node quantum random network. Our integrated platform exploits probabilistic photon-pair sources and coherent control of path modes to prepare a structured quantum subgraph through local transformations and postselection, operating in a threshold regime that differs from classical random networks. We verify that the subgraph state exhibits genuine high-dimensional multipartite entanglement across the nodes, providing experimental evidence that quantum entanglement enables connectivity structures beyond classical accessibility.

quant-ph↗