arXiv subjects

Kai Zhou

Publications and source records attributed to Kai Zhou.

At least 19 recordsLinked to original sources

arXiv2026

Singular Cholesky Fibers over Finite Fields

For a finite field $\F_q$, consider the triangular Cholesky map $\Gamma_{n,q}(U)=U^TU$ from upper triangular matrices to symmetric matrices. Generalized Cholesky theory describes the regular locus on which all leading principal minors are nonzero, but it does not determine the multiplicities or root ranks in a fiber over a singular target. We develop a fixed-target fiber theory for this singular boundary and answer three questions posed by Cooper and Whitlatch. Our principal results concern the zero fiber. Over $\F_2$ we construct an explicit, invertible, rank-preserving recursive bijection between square-zero upper triangular matrices and upper triangular matrices satisfying $U^TU=0$. Over every finite field we determine the entire rank distribution of this zero fiber: in even characteristic the square-zero recurrence, and hence the rank-refined equinumerosity, persists over every $\F_{2^e}$, whereas in odd characteristic its failure is measured by an explicit quadratic-character correction. We place these results in a uniform framework by proving an exact first-pivot recursion for the fiber cardinality over an arbitrary symmetric target. Its regular specialization gives the constant fiber sizes on leading-principal-minor cones, while its zero-pivot branch explains the new singular behavior. For binary diagonal targets we further compress the rank-refined count to $O(n^2)$ integer-arithmetic transitions and show that it depends on the order, not merely the number, of the diagonal entries. Executable implementations and exhaustive low-order checks are collected in an appendix.

math.RA↗

arXiv2026

DSPrompt: Dynamic Soft Prompt Defense Against M-RAG Corruption

Multimodal Retrieval Augmented Generation (M-RAG) is increasingly vulnerable to adversarial attacks where malicious data are crafted to produce embeddings that align with benign entries in the vector space, deceiving retrieval and inducing harmful outputs. Existing defenses primarily operate at query time, relying on auxiliary detectors, similarity re-ranking, or feature-consistency checks. However, these approaches suffer from non-trivial inference overhead, generalize poorly to unseen attack strategies, and often assume specific attack distributions. To address this, we propose DSPrompt, a Dynamic Soft Prompt defense framework that directly reshapes the retriever's embedding semantics, without modifying the retrieval pipeline. It inserts few learnable soft prompts into each layer of the visual and textual encoders of a frozen retriever, utilizing a shallow-to-deep length schedule that is adaptive to the capacity in the model layers. These prompts are trained under a dynamic min-max scheme: an online multimodal attacker continually crafts hard adversarial documents against the current retriever, while the defender is updated to push such documents out of the top-k while preserving the ranking and diversity of benign evidence. Because the defended encoder can be pre-computed and indexed exactly as in standard dense retrieval, DSPrompt incurs no additional per-query optimization and introduces fewer than 1% additional parameters. Extensive experiments across four benchmarks and three representative poisoning attacks show that DSPrompt substantially reduces the attack success rate and poison retrieval rate while maintaining near-lossless retrieval utility and generation fidelity, consistently outperforming existing defense baselines at a fraction of their computational cost.

cs.CR↗

arXiv2026

Phase-continuous comparison of three all-optical time scales over 20 days

Optical frequency standards have progressed rapidly over the past two decades, leading to the anticipated redefinition of the SI second by an optical frequency. However, time scales have not yet significantly improved despite this development because they are still fully reliant on rf flywheel oscillators, mostly hydrogen masers, which impose a performance limit related to incompletely sampled noise known as the Dick effect. To best benefit from the exceptional stability and accuracy of optical frequency standards, time scales must employ optical flywheels with orders-of-magnitude better short-term (<$10^4$ s) stability than masers. Here, we introduce three optical flywheel oscillators (two cryogenic silicon cavities and one iodine optical clock) with superior short-term stability than hydrogen masers and long-term stability on par with masers. Steering each optical flywheel with a high-uptime Sr optical frequency standard generates three parallel all-optical time scales with continuous operation over >20 days. When compared with each other, these all-optical time scales achieve <$10^{-16}$ relative instability after just a few days of averaging. During typical steering gaps of ~6 hours, the accumulated time difference is ~20 ps, leading to the total time difference of <100 ps over the full measurement period. With the proliferation of long-distance optical fiber links and commercialization of optical flywheels and frequency standards, we anticipate all-optical time scales to be the future of timekeeping.

physics.atom-ph↗

arXiv2026

The Greedy Advantage in Finite-Horizon Bandits

Organizations increasingly rely on sequential experimentation to improve decision-making. While the multi-armed bandit literature has developed algorithms with strong asymptotic regret guarantees, many practical applications operate over finite and externally imposed horizons. Motivated by the finite-horizon setting, we develop a class of regularized greedy algorithms for multi-armed Bernoulli bandits. We derive the first finite-horizon regret envelopes for regularized greedy bandits, showing that finite-horizon regret decomposes into transient exploration costs and a suboptimal convergence term that decays exponentially with the regularization strength. This characterization yields principled calibration rules for the regularization parameters and, as a limiting case, sharper regret guarantees for the classical greedy policy. Across extensive numerical experiments, calibrated regularized greedy policies consistently match or outperform state-of-the-art algorithms. These results suggest that regularized greedy policies can provide an effective approach for finite-horizon bandit problems.

stat.ML↗

arXiv2026

JD Oxygen AI Item Center (Oxygen AIIC) V1: An Industrial-Scale LLM/VLM-Centric Solution for Item Understanding, Management, and Applications

JD$.$com, one of the world's largest e-commerce platforms, serves over 700 million active users and millions of merchants, with a catalog of tens of billions of SKUs. At this scale, high-quality, structured item knowledge underpins a better consumer experience, lower management costs, and higher operational efficiency-yet producing and serving it poses three industrial-scale challenges: fast-emerging concepts, high-quality knowledge production for massive SKUs, and diverse downstream requirements. To address these challenges, we present the JD Oxygen AI Item Center (Oxygen AIIC), an industrial-scale platform built on LLMs/VLMs for item-knowledge production and service. Oxygen AIIC is built around four core pillars: (i) ontology engineering driven by efficient human-AI collaboration, which supports the dynamic evolution and agile expansion of an ontology with millions of entries; (ii) a "Semantic Search then Discrimination"(S2D) knowledge identification architecture that, combined with throughput improvement strategies, enables scalable, extensible, and high-throughput AI Item Library production for tens of billions of SKUs; (iii) self-evolving item-understanding LLMs/VLMs that improve in a stable and controllable manner, enabling knowledge production with 94.2% precision and 82.8% recall; and (iv) a unified item tunnel that serves as the data and service hub. Oxygen AIIC now covers tens of thousands of JD categories and processes hundreds of millions of item updates per day on Huawei Ascend NPUs. It has accumulated hundreds of billions of item-knowledge assets. Deployed across core business scenarios-including search, recommendation, operations, category planning-Oxygen AIIC has delivered measurable gains at scale. Search-traffic coverage reaches 80.4%, item-information quality issues drop by 37%, the automated fill rate of core attributes during item listing exceeds 80%.

cs.AI↗

arXiv2026

AdaThinking-E: One-Token Entropy Regulation for Adaptive Thinking

Multimodal large language models have demonstrated strong document reasoning capabilities by incorporating explicit thinking processes. While this capability significantly improves performance on challenging tasks, current models apply such deep reasoning uniformly to all questions, resulting in unnecessary computational overhead for simple task. This not only degrades user experience but also negatively impact accuracy on benchmark datasets. We identify the critical need for adaptive thinking mechanisms that can intelligently determine when to engage reasoning based on question complexity. To address this, we propose AdaThinking-E, a novel reinforcement learning framework that learns adaptive thinking through one-token entropy regulation. Our key insight is that model confidence in the decision to engage thinking (or not) can be quantified through entropy analysis of the predicted probability distribution at critical decision tokens. This observation motivates our entropy-governed reward mechanism: the training process naturally transitions from high-entropy exploration, where the model experiments with different thinking strategies, to low-entropy convergence with confident, generalizable decision-making policies. Crucially, this approach enables models to intrinsically discover when to think without requiring manual intervention or external difficulty labels. Extensive experiments demonstrate that our approach enables models to be both accurate on complex problems and efficient on simple ones across diverse document tasks.

cs.CL↗

arXiv2026

REDI-Match: Rotation-Equivariant Distillation for Efficient and Robust Dense Matching

Vision Foundation Models (VFMs) have significantly advanced dense feature matching, yet severe in-plane rotation remains a critical challenge. Existing solutions face a fundamental dilemma: data-driven methods require inefficient parameter scaling to implicitly learn rotations, whereas strictly equivariant networks lack the semantic capacity of modern VFMs. Consequently, current frameworks typically freeze VFMs and shift the entire burden of rotation generalization to the downstream decoder. To break this architectural bottleneck, we propose REDI-Match, an efficient framework driven by a novel Rotation-Equivariant Distillation (REDI) paradigm. Instead of relying on rotation data augmentation to establish rotational correspondences, REDI distills the non-equivariant semantic representations of a VFM into a lightweight, strictly rotation-equivariant encoder, leveraging an equivariant geometric architecture to constrain robust high-dimensional semantics. To fully exploit these features, we equip the decoder with an entropy-driven spatial alignment module. By evaluating discrete rotation hypotheses, this mechanism explicitly locks onto the canonical coordinate system, eliminating global ambiguity before continuous refinement. Extensive experiments demonstrate that REDI-Match establishes a new state-of-the-art (SOTA) across multiple benchmarks. Notably, it achieves a 13.89% absolute pose accuracy improvement on the highly challenging SatAst dataset while operating 1.9x faster than the current SOTA (RoMa v2), enabling real-time inference (~41 FPS) on a single RTX 4090 GPU. Code: https://github.com/YinjiGe/REDI-Match.

cs.CV↗

arXiv2026

ViTexQA: A Multi-Frame Temporal Perception Dataset for Video Text Question Answering

Despite remarkable progress in multimodal understanding, current MLLMs still exhibit limitations in video text understanding, particularly when semantics emerge through the integration of temporally distributed textual cues across multiple frames. This perception challenge fundamentally differs from static image text understanding, yet existing datasets fail to capture: the vast majority of questions remain answerable from single frames, inadequately reflecting real-world video text comprehension demands. To address this, we present ViTexQA, a large-scale video-text QA dataset, and FrameThinker for robust multi-frame temporal reasoning. We build ViTexQA via a quality-controlled Chain-of-Thought (CoT) annotation pipeline boosted with temporal constraints; all its QA pairs demand cross-frame text fusion to solve, enforcing true temporal reliance. FrameThinker adopts two-stage training for explicit temporal modeling: CoT-Guided Supervised Fine-Tuning (SFT) generates frame-aware reasoning chains, followed by Temporally-grounded Reinforcement Learning (RL) optimized with multi-frame coherence rewards. Evaluations show our method outperforms SOTA baselines on ViTexQA, lifting ROUGE-L by 6.3%.

cs.CV↗

arXiv2026

Stochastic Path Sampler For Lattice Field Theory

In lattice field theory, target distributions are known only up to normalization, (\tilde{\pi}(\phi)\propto e^{-S(\phi)}), while the partition function is intractable. Markov chain Monte Carlo simulations often become inefficient near phase transitions or the continuum limit due to critical slowing down. In this work, we propose a novel sampler based on nonequilibrium thermodynamics, called Stochastic Path Sampler (SPS), which can generate configurations for the unnormalized target distribution without requiring training data. The central idea of SPS is to establish a trajectory-level balance for learnable forward and backward stochastic dynamics between two equilibrium states, namely the prior and target distributions. This is achieved by minimizing the path-space variational free energy, equivalently an entropy-production upper bound, defined by the log-ratio of forward and auxiliary backward trajectory measures, thereby enhancing the reversibility of the forward and backward processes. The learned forward process provides independent proposals, which are subsequently corrected by an extended-space Independence Metropolis--Hastings step. In two-dimensional (\phi^4) theory, we demonstrate that our neural sampler can achieve the same sampling quality as HMC but with a much shorter autocorrelation time in the critical region. This sampler offers a stochastic-quantization-inspired route to data-free proposal construction for lattice field theory by leveraging a variational free-energy principle derived from path-space irreversibility.

hep-lat↗

arXiv2026

Toward Polymorphic Backdoor against Semantic Communication via Intensity-Based Poisoning

Semantic Communication (SC) backdoor attacks aim to utilize triggers to manipulate the system into producing predetermined outputs via backdoored shared knowledge. Current SC backdoors adopt monomorphic paradigms with single attack target, which suffers from limited attack diversity, efficiency, and flexibility in heterogeneous downstream scenarios. To overcome the limitations, we propose SemBugger, a polymorphic SC backdoor. By dynamically adjusting the trigger intensity, SemBugger finely-grained controls over the SC knowledge to generate diverse malicious results from the system. Specifically, SemBugger is realized through a multi-effect poisoning-training framework. It introduces graded-intensity triggers to poison training data and optimizes SC systems with hierarchical malicious loss. The trained system's knowledge dynamically adapts to trigger intensity in inputs to yield target outputs, all while preserving transmission fidelity for benign samples. Moreover, to augment SC security, we propose a provable robustness defense that resists SemBugger's homogeneous attacks through a controlled noise mechanism. It operates via strategically adding noise in SC inputs, and we formally provide a theoretical lower bound on the defense efficacy. Experiments across diverse SC models and benchmark datasets indicate that SemBugger attains high attack efficacy while maintaining the regular functionality of SC systems. Meanwhile, the designed defense effectively neutralizes SemBugger attacks.

cs.CR↗

arXiv2026

Instance-level Visual Active Tracking with Occlusion-Aware Planning

Visual Active Tracking (VAT) aims to control cameras to follow a target in 3D space, which is critical for applications like drone navigation and security surveillance. However, it faces two key bottlenecks in real-world deployment: confusion from visually similar distractors caused by insufficient instance-level discrimination and severe failure under occlusions due to the absence of active planning. To address these, we propose OA-VAT, a unified pipeline with three complementary modules. First, a training-free Instance-Aware Offline Prototype Initialization aggregates multi-view augmented features via DINOv3 to construct discriminative instance prototypes, mitigating distractor confusion. Second, an Online Prototype Enhancement Tracker enhances prototypes online and integrates a confidence-aware Kalman filter for stable tracking under appearance and motion changes. Third, an Occlusion-Aware Trajectory Planner, trained on our new Planning-20k dataset, uses conditional diffusion to generate obstacle-avoiding paths for occlusion recovery. Experiments demonstrate OA-VAT achieves 0.93 average SR on UnrealCV (+2.2% vs. SOTA TrackVLA), 90.8% average CAR on real-world datasets (+12.1% vs. SOTA GC-VAT), and 81.6% TSR on a DJI Tello drone. Running at 35 FPS on an RTX 3090, it delivers robust, real-time performance for practical deployment.

cs.CV↗

arXiv2026

HalluClear: Diagnosing, Evaluating and Mitigating Hallucinations in GUI Agents

While progress in GUI agents has been largely driven by industrial-scale training, ungrounded hallucinations often trigger cascading failures in real-world deployments.Unlike general VLM domains, the GUI agent field lacks a hallucination-focused suite for fine-grained diagnosis, reliable evaluation, and targeted mitigation.To bridge this gap, we introduce HalluClear, a comprehensive suite for hallucination mitigation in GUI agents as a complement to computation-intensive scaling. HalluClear comprises: (1) a GUI-specific hallucination taxonomy derived from empirical failure analysis; (2) a calibrated three-stage evaluation workflow which enhances VLM-as-a-judge reliability via expert-annotated benchmarking and ensemble credibility estimation; and (3) a mitigation scheme based on closed-loop structured reasoning, enabling lightweight continual post-training with cold-start initialization for both generalist and GUI-specialist agents. Experiments across representative agents and public benchmarks demonstrate that post-training on only 9K samples within our suite can significantly reduce hallucinations, thereby improving grounding and action fidelity, offering a compute-efficient pathway to robust GUI automation.

cs.AI↗

arXiv2026

Training-Free Test-Time Contrastive Learning for Large Language Models

Large language models (LLMs) demonstrate strong reasoning capabilities, but their performance often degrades under distribution shift. Existing test-time adaptation (TTA) methods rely on gradient-based updates that require white-box access and need substantial overhead, while training-free alternatives are either static or depend on external guidance. In this paper, we propose Training-Free Test-Time Contrastive Learning TF-TTCL, a training-free adaptation framework that enables a frozen LLM to improve online by distilling supervision from its own inference experiences. Specifically, TF-TTCL implements a dynamic "Explore-Reflect-Steer" loop through three core modules: 1) Semantic Query Augmentation first diversifies problem views via multi-agent role-playing to generate different reasoning trajectories; 2) Contrastive Experience Distillation then captures the semantic gap between superior and inferior trajectories, distilling them into explicit textual rules; and 3) Contextual Rule Retrieval finally activates these stored rules during inference to dynamically steer the frozen LLM toward robust reasoning patterns while avoiding observed errors. Extensive experiments on closed-ended reasoning tasks and open-ended evaluation tasks demonstrate that TF-TTCL consistently outperforms strong zero-shot baselines and representative TTA methods under online evaluation. Code is available at https://github.com/KevinSCUTer/TF-TTCL.

cs.CL↗

arXiv2026

Sensitivity of Neutron Star Observables to Transition Density in Hybrid Equation-of-State Models

We investigate how the transition density $\rho_{tr}$ affects hybrid constructions of the neutron-star equation of state (EoS) in which a nucleonic description at low densities is matched to a model-agnostic high-density extension based on a speed-of-sound parametrization. Using four representative nucleonic models--Taylor expansion, $\frac{n}{3}$ expansion, Skyrme, and relativistic mean-field--built from identical nuclear matter parameters, we isolate the impact of the low-density EoS and the transition density on neutron star observables. We find that, within the present smooth-matching prescription, neutron star properties such as radii and tidal deformabilities retain significant sensitivity to the choice of low-density EoS for commonly adopted transition densities around $\rho_{tr} \approx 2\rho_0$, even when the same high-density parametrization is employed. This residual dependence arises from differences in the matching conditions at $\rho_{tr}$, which propagate into the high-density extension, so different low-density inputs lead to different effective high-density EoSs. These findings are robust across two distinct speed-of-sound parametrizations. Quantitatively, the model spread in radius and tidal deformability at $1.4\,M_\odot$ exceeds the current observational uncertainty by factors of $\sim 1.8$ and $\sim 1.4$ at $\rho_{\mathrm{tr}} \approx 2\rho_0$, whereas these factors reduce to $\sim 1.05$ and $\sim 0.4$ at $\rho_{\mathrm{tr}} = \rho_0$. Lowering the transition density, therefore, systematically diminishes the spread among models and leads to more consistent predictions. Our results demonstrate that the widely used choice $\rho_{tr} \approx 2\rho_0$ does not guarantee model independence in hybrid EoS constructions, and should be treated as an explicit source of systematic uncertainty when inferring dense matter properties from neutron star observations.

nucl-th↗

arXiv2026

Unified Extraction of In-Medium Heavy Quark Potentials from RHIC to LHC Energies via Deep Learning

We use deep learning under Bayesian perspective to quantitatively extract the in-medium heavy quark (HQ) potential from bottomonium nuclear modification factors ($R_{AA}$) measured across multiple heavy ion collision systems at the Large Hadron Collider (LHC) and the Relativistic Heavy-Ion Collider (RHIC). The in-medium HQ potential, comprising both a real and imaginary part, is parameterized and incorporated into a time-dependent Schr\"odinger equation to model the wave function evolution of $b\bar{b}$ dipoles within a hydrodynamically evolving hot QCD medium. We construct Convolutional Neural Networks (CNNs) to capture the non-linear correspondence between the heavy quark potential $V(T,r)$ and the bottomonium $R_{AA}$ for Pb-Pb collisions at 5.02 TeV and 2.76 TeV, and Au-Au collisions at 200 GeV. Training datasets are generated by sampling the potential parameters and are further augmented using Principal Component Analysis (PCA) and Gaussian Process Regression (GPR). After validating the stability and correctness of the CNNs, we employ Stochastic Gradient Langevin Dynamics (SGLD) to perform a simultaneous Bayesian inverse extraction of the optimal potential parameters and their posterior distributions using experimental data of bottomonium $R_{AA}$ in both LHC and RHIC energies. Our joint multi-energy extraction suggests that, within the present parametrization and hydrodynamic background, the real part of the in-medium potential remains close to the vacuum Cornell form, corresponding to a relatively weak screened Debye mass across RHIC to LHC energies. By contrast, the imaginary part is more strongly constrained by the data and provides the dominant contribution to bottomonium suppression from RHIC to LHC energies.

nucl-th↗

arXiv2026

Deep Hierarchical Knowledge Loss for Fault Intensity Diagnosis

Fault intensity diagnosis (FID) plays a pivotal role in intelligent manufacturing while neglecting dependencies among target classes hinders its practical deployment. This paper introduces a novel and general framework with deep hierarchical knowledge loss (DHK) to achieve hierarchical consistent representation and prediction. We develop a novel hierarchical tree loss to enable a holistic mapping of same-attribute classes, leveraging tree-based positive and negative hierarchical knowledge constraints. We further design a focal hierarchical tree loss to enhance its extensibility and devise two adaptive weighting schemes based on tree height. In addition, we propose a group tree triplet loss with hierarchical dynamic margin by incorporating hierarchical group concepts and tree distance to model boundary structural knowledge across classes. The joint two losses significantly improve the recognition of subtle faults. Extensive experiments are performed on four real-world datasets from various industrial domains (three cavitation datasets from SAMSON AG and one publicly available dataset) for FID, all showing superior results and outperforming recent state-of-the-art FID methods.

eess.AS↗

arXiv2026

HandX: Scaling Bimanual Motion and Interaction Generation

Synthesizing human motion has advanced rapidly, yet realistic hand motion and bimanual interaction remain underexplored. Whole-body models often miss the fine-grained cues that drive dexterous behavior, finger articulation, contact timing, and inter-hand coordination, and existing resources lack high-fidelity bimanual sequences that capture nuanced finger dynamics and collaboration. To fill this gap, we present HandX, a unified foundation spanning data, annotation, and evaluation. We consolidate and filter existing datasets for quality, and collect a new motion-capture dataset targeting underrepresented bimanual interactions with detailed finger dynamics. For scalable annotation, we introduce a decoupled strategy that extracts representative motion features, e.g., contact events and finger flexion, and then leverages reasoning from large language models to produce fine-grained, semantically rich descriptions aligned with these features. Building on the resulting data and annotations, we benchmark diffusion and autoregressive models with versatile conditioning modes. Experiments demonstrate high-quality dexterous motion generation, supported by our newly proposed hand-focused metrics. We further observe clear scaling trends: larger models trained on larger, higher-quality datasets produce more semantically coherent bimanual motion. Our dataset is released to support future research.

cs.CV↗

arXiv2026

ScoutAttention: Efficient KV Cache Offloading via Layer-Ahead CPU Pre-computation for LLM Inference

Large language models encounter critical GPU memory capacity constraints during long-context inference, where KV cache memory consumption severely limits decode batch sizes. While existing research has explored offloading KV cache to DRAM, these approaches either demand frequent GPU-CPU data transfers or impose extensive CPU computation requirements, resulting in poor GPU utilization as the system waits for I/O operations or CPU processing to complete. We propose ScoutAttention, a novel KV cache offloading framework that accelerates LLM inference through collaborative GPU-CPU attention computation. To prevent CPU computation from bottlenecking the system, ScoutAttention introduces GPU-CPU collaborative block-wise sparse attention that significantly reduces CPU load. Unlike conventional parallel computing approaches, our framework features a novel layer-ahead CPU pre-computation algorithm, enabling the CPU to initiate attention computation one layer in advance, complemented by asynchronous periodic recall mechanisms to maintain minimal CPU compute load. Experimental results demonstrate that ScoutAttention maintains accuracy within 2.4% of baseline while achieving 2.1x speedup compared to existing offloading methods.

cs.LG↗