arXiv subjects

Kai Zhou

Publications and source records attributed to Kai Zhou.

At least 37 records · Page 2Linked to original sources

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

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 Schrödinger 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

Deep learning approaches to extract nuclear deformation parameters from initial-state information in heavy-ion collisions

The deformation of heavy nuclei leaves characteristic imprints on the initial conditions of relativistic heavy-ion collisions. However, event-by-event fluctuations make the quantitative extraction of this information challenging. This study examines the identifiability of the quadrupole ($β_2$) and hexadecapole ($β_4$) deformation parameters from nucleon configurations sampled from a deformed Woods-Saxon distribution commonly used in initial-state modeling of heavy-ion collisions. As a baseline, we first establish an upper bound on the "intrinsic identifiability" of deformation information at the most microscopic level by constructing permutation-invariant point-cloud networks under controlled multi-event grouping. We then extend the analysis to the more realistic initial entropy-density profiles generated by the TRENTo model, where both standard regression and simulation-based inference (SBI) with conditional normalizing flows are employed to reconstruct the deformation parameters from ensembles of event images supplemented with global attributes. Multi-event averaging is found to be essential in this setting for suppressing stochastic fluctuations and revealing the underlying deformation information. While standard regression efficiently captures the central trends of deformation through point estimates, SBI provides calibrated posterior distributions, offering a more complete and robust characterization of uncertainty. Collectively, our results demonstrate that deformation information is effectively encoded in the initial state and becomes increasingly identifiable with sufficient ensemble averaging, laying a solid foundation for future extensions toward more complete dynamical modeling and final-state observables.

nucl-th↗

arXiv2026

Neural-Network Holographic Model of the QCD Phase Transition under Lattice and HRG Constraints

Within a neural-network-based holographic framework, we incorporate lattice QCD (LQCD) and Hadron Resonance Gas (HRG) data to train the model and predict the location of the QCD critical endpoint (CEP). The training dataset consists of the entropy density, baryon number susceptibility, and baryon density. The metric warp factor $A(z)$ and the gauge kinetic function $f(z)$ are parameterized by neural networks and determined through the training procedure. The resulting model reproduces the equation of state at vanishing chemical potential in good agreement with both LQCD and HRG data. Extending the analysis to finite chemical potential, we solve the equations of motion and obtain thermodynamic observables consistent with LQCD results at finite density. After incorporating the HRG constraints, the predicted position of the CEP shifts toward larger chemical potentials compared to recent studies. We further employ symbolic regression to derive analytic expressions for $A(z)$ and $f(z)$, providing convenient functional forms for future phenomenological applications. Finally, we perform a data-driven validation using synthetic thermodynamic data generated from an existing analytical holographic model. The neural-network framework reproduces the corresponding CEP location with good accuracy, showing close agreement within numerical uncertainties.

hep-ph↗

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↗

arXiv2026

VeRA+: Vector-Based Lightweight Digital Compensation for Drift-Resilient RRAM In-Memory Computing

RRAM-based in-memory computing (IMC) offers high energy efficiency but suffers from conductance drift that severely degrades long-term accuracy. Existing approaches including retraining, noise-aware training, and Batch Normalization (BN)-based calibration either require RRAM rewriting, demand large storage overhead, or rely on online correction. We propose VeRA+, a lightweight drift compensation framework that reuses shared projection matrices and introduces only two compact drift-specific vectors per drift level. A drift-aware scheduling algorithm offline-trains a small set of VeRA+ parameters and selects the appropriate set over time without any on-chip retraining or data replay. VeRA+ preserves up to 99.77% of the drift-free accuracy after ten years of simulated drift and reduces storage overhead by more than three orders of magnitude compared with BN-based calibration. To validate VeRA+ under realistic device behavior, we extract one-week drift statistics from measurements on our fabricated 1T1R RRAM devices and use them to simulate realistic drifted weights. Under these measured drift conditions, VeRA+ achieves accuracy close to the drift-free baseline, providing an efficient and practical solution for long-term drift resilience in RRAM-IMC.

cs.AR↗

arXiv2026

PositionOCR: Augmenting Positional Awareness in Multi-Modal Models via Hybrid Specialist Integration

In recent years, Multi-modal Large Language Models (MLLMs) have achieved strong performance in OCR-centric Visual Question Answering (VQA) tasks, illustrating their capability to process heterogeneous data and exhibit adaptability across varied contexts. However, these MLLMs rely on a Large Language Model (LLM) as the decoder, which is primarily designed for linguistic processing, and thus inherently lacks the positional reasoning required for precise visual tasks, such as text spotting and text grounding. Additionally, the extensive parameters of MLLMs necessitate substantial computational resources and large-scale data for effective training. Conversely, text spotting specialists achieve state-of-the-art coordinate predictions but lack semantic reasoning capabilities. This dichotomy motivates our key research question: Can we synergize the efficiency of specialists with the contextual power of LLMs to create a positionally-accurate MLLM? To overcome these challenges, we introduce PositionOCR, a parameter-efficient hybrid architecture that seamlessly integrates a text spotting model's positional strengths with an LLM's contextual reasoning. Comprising 131M trainable parameters, this framework demonstrates outstanding multi-modal processing capabilities, particularly excelling in tasks such as text grounding and text spotting, consistently surpassing traditional MLLMs.

cs.CV↗

arXiv2026

Bayesian Inference of Neutron Star Properties in $f(Q)$ Gravity Using NICER Observations

In this work, we investigate neutron stars (NSs) in the strong field regime within the framework of symmetric teleparallel $f(Q)$ gravity, considering three representative models: linear, logarithmic, and exponential. While Bayesian studies of NS observations are well established in general relativity and curvature based modified gravity theories, such analyses in $f(Q)$ gravity remain largely unexplored. For the first time we perform a Bayesian inference analysis by confronting theoretical NS mass-radius predictions with NICER observations of PSR J0030+0451, PSR J0740+6620, PSR J0437+4715, and PSR J0614+3329 in the background of nonmetricity based gravity. The dense matter equation of state is fixed to DDME2 in order to isolate the effects of modified gravity on NS structure. Our results show that the exponential $f(Q)$ model is statistically preferred over the linear and logarithmic cases, as confirmed by Bayes factor comparisons, and exhibits well-constrained. For this model, we obtain a radius and tidal deformability at $1.4\,M_\odot$ of $R_{1.4} = 11.27^{+0.53}_{-0.36}\,\mathrm{km}$ and $Λ_{1.4} = 156.95^{+84.02}_{-41.73}$, respectively, consistent with current observational constraints. Remarkably, all three constrained models predict maximum neutron star masses reaching $M_{\max} \simeq 2.98\,M_{\odot}$, with the $95\%$ confidence regions extending into the lower mass gap ($\sim 2.5$--$5\,M_{\odot}$). This mass-gap prediction emerges naturally from the Bayesian-constrained parameter space. These results highlight the potential of NSs as powerful probes of symmetric teleparallel gravity in the strong field regime.

gr-qc↗

arXiv2026

Context Forcing: Consistent Autoregressive Video Generation with Long Context

Recent approaches to real-time long video generation typically employ streaming tuning strategies, attempting to train a long-context student using a short-context (memoryless) teacher. In these frameworks, the student performs long rollouts but receives supervision from a teacher limited to short 5-second windows. This structural discrepancy creates a critical \textbf{student-teacher mismatch}: the teacher's inability to access long-term history prevents it from guiding the student on global temporal dependencies, effectively capping the student's context length. To resolve this, we propose \textbf{Context Forcing}, a novel framework that trains a long-context student via a long-context teacher. By ensuring the teacher is aware of the full generation history, we eliminate the supervision mismatch, enabling the robust training of models capable of long-term consistency. To make this computationally feasible for extreme durations (e.g., 2 minutes), we introduce a context management system that transforms the linearly growing context into a \textbf{Slow-Fast Memory} architecture, significantly reducing visual redundancy. Extensive results demonstrate that our method enables effective context lengths exceeding 20 seconds -- 2 to 10 times longer than state-of-the-art methods like LongLive and Infinite-RoPE. By leveraging this extended context, Context Forcing preserves superior consistency across long durations, surpassing state-of-the-art baselines on various long video evaluation metrics.

cs.CV↗

arXiv2026

The neural networks with tensor weights and emergent fermionic Wick rules in the large-width limit

In this paper, we study complex-valued neural network (CVNNs) with tensor-valued hidden-to-output weights within the framework of neural-network quantum field theory (NN-QFT). For standard CVNNs with scalar weights, we derive the generating functional and identify the exact Gaussian process that arises in the infinite-width limit, together with its associated effective quantum state. When the last-layer weights are promoted to Clifford-algebra-valued tensors, the network output becomes complex matrix-valued, and a fermion-like sign structure in the large-width correlation functions of the network output is induced. We show that, in the infinite-width limit, correlators with equal numbers of $f^†$ and $f$ obey fermionic Wick rules and can be written as determinants built from a scalar Euclidean kernel $S(x,y)=\langle f^†(x)f(y)\rangle$. This provides a sign-structured extension of NN-QFT at the level of Euclidean correlators and Feynman rules, even though a microscopic Grassmann path integral representation for the network parameters has not yet been constructed. Our analysis thus pushes the NN-QFT correspondence beyond purely bosonic Gaussian fields and suggests a possible route to encoding fermion-like symmetries in neural architectures for QFT correspondence.

hep-th↗

arXiv2026

Physics-Conditioned Diffusion Models for Lattice Gauge Theory

We develop diffusion models for simulating lattice gauge theories, where stochastic quantization is explicitly incorporated as a physical condition for sampling. We demonstrate the applicability of this novel sampler to U(1) gauge theory in two spacetime dimensions and find that a model trained at a small inverse coupling constant can be extrapolated to larger inverse coupling regions without encountering the topological freezing problem. Additionally, the trained model can be employed to sample configurations on different lattice sizes without requiring further training. The exactness of the generated samples is ensured by incorporating Metropolis-adjusted Langevin dynamics into the generation process. Furthermore, we demonstrate that this approach enables more efficient sampling of topological quantities compared to traditional algorithms such as Hybrid Monte Carlo and Langevin simulations.

hep-lat↗

arXiv2026

Learning the relations between neutron star and nuclear matter properties with symbolic regression

The equation of state (EOS) of dense matter in neutron stars (NSs) remains uncertain, particularly at supra-nuclear densities where complex nuclear interactions and the potential presence of exotic matter, like hyperons, come into play. The complex relationships existing between nuclear matter and neutron star properties are investigated. The focus is on their nonlinearities and interdependencies. In our analysis, we apply a machine learning algorithm known as symbolic regression, paired with principal component analysis, to datasets generated from Bayesian inference over relativistic mean-field models. A systematic Principal Component Analysis has allowed to break down the percentage contribution of each element or feature in the relationships obtained. This study examines two main models (datasets): the NL model, which includes nucleonic degrees of freedom; and the NL-hyp model, which includes hyperons in addition to nucleons. Our analysis confirms a robust correlation between the tidal deformability of a 1.4 $M_\odot$ neutron star and $β$-equilibrium pressure at twice the nuclear saturation density. This correlation remains once hyperons are included. The contribution of the different nuclear matter properties at saturation to the radius and tidal deformability was calculated. It was shown that the isovector properties have the largest impact, with a contribution of about 90\%. We also studied the relationship between the proton fraction at different densities and various symmetry energy parameters defined at saturation density. For the hyperon data set, we took into account the effects of the negatively charged hyperon $Ξ$ in order to recover the relationships. Our study reveals the individual impact of various symmetry energy parameters on proton fractions at different densities.

nucl-th↗

arXiv2026

A deep learning approach for predicting multiple observables in Au+Au collisions at RHIC

We develop a neural network model, based on the processes of high-energy heavy-ion collisions, to study and predict several experimental observables in Au+Au collisions. We present a data-driven deep learning framework for predicting multiple bulk observables in Au+Au collisions at RHIC energies. A single neural network is trained exclusively on experimental measurements of charged-particle pseudorapidity density distributions, transverse-momentum spectra and elliptic flow coefficients over a broad range of collision energies and centralities. The network architecture is inspired by the stages of a heavy-ion collision, from the quark-gluon plasma to chemical and kinetic freeze-out, and employs locally connected hidden layers and a structured input design that encodes basic geometric and kinematic features of the system. We demonstrate that these physics-motivated choices significantly improve test performance compared to purely fully connected baselines. The trained model is then used to predict the above observables at collision energies not yet explored experimentally at RHIC, and the results are validated using the energy dependence of the total charged-particle multiplicity per participant pair as well as comparisons to a CLVisc hydrodynamic calculation with TRENTo initial conditions. Our findings indicate that such physics-guided neural networks can serve as efficient surrogates to fill critical data gaps at RHIC and to support further phenomenological studies of QGP properties.

nucl-th↗