arXiv subjects

Yixin Zhang

Publications and source records attributed to Yixin Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Speech-Aware Long Context Pruning and Integration for Contextualized Automatic Speech Recognition

Automatic speech recognition (ASR) systems have achieved remarkable performance in common conditions but often struggle to leverage long-context information in contextualized scenarios that require domain-specific knowledge, such as conference presentations. This challenge arises primarily due to constrained model context windows and the sparsity of relevant information within extensive contextual noise. To solve this, we propose the SAP$^{2}$ method, a novel framework that dynamically prunes and integrates relevant contextual keywords in two stages. Specifically, each stage leverages our proposed Speech-Driven Attention-based Pooling mechanism, enabling efficient compression of context embeddings while preserving speech-salient information. Experimental results demonstrate state-of-the-art performance of SAP$^{2}$ on the SlideSpeech and LibriSpeech datasets, achieving word error rates (WER) of 7.71% and 1.12%, respectively. On SlideSpeech, our method notably reduces biased keyword error rates (B-WER) by 41.1% compared to non-contextual baselines. SAP$^{2}$ also exhibits robust scalability, consistently maintaining performance under extensive contextual input conditions on both datasets.

cs.CL↗

arXiv2025

When Alignment Fails: Multimodal Adversarial Attacks on Vision-Language-Action Models

Vision-Language-Action models (VLAs) have recently demonstrated remarkable progress in embodied environments, enabling robots to perceive, reason, and act through unified multimodal understanding. Despite their impressive capabilities, the adversarial robustness of these systems remains largely unexplored, especially under realistic multimodal and black-box conditions. Existing studies mainly focus on single-modality perturbations and overlook the cross-modal misalignment that fundamentally affects embodied reasoning and decision-making. In this paper, we introduce VLA-Fool, a comprehensive study of multimodal adversarial robustness in embodied VLA models under both white-box and black-box settings. VLA-Fool unifies three levels of multimodal adversarial attacks: (1) textual perturbations through gradient-based and prompt-based manipulations, (2) visual perturbations via patch and noise distortions, and (3) cross-modal misalignment attacks that intentionally disrupt the semantic correspondence between perception and instruction. We further incorporate a VLA-aware semantic space into linguistic prompts, developing the first automatically crafted and semantically guided prompting framework. Experiments on the LIBERO benchmark using a fine-tuned OpenVLA model reveal that even minor multimodal perturbations can cause significant behavioral deviations, demonstrating the fragility of embodied multimodal alignment.

cs.CV↗

arXiv2025

CureAgent: A Training-Free Executor-Analyst Framework for Clinical Reasoning

Current clinical agent built on small LLMs, such as TxAgent suffer from a \textit{Context Utilization Failure}, where models successfully retrieve biomedical evidence due to supervised finetuning but fail to ground their diagnosis in that information. In this work, we propose the Executor-Analyst Framework, a modular architecture that decouples the syntactic precision of tool execution from the semantic robustness of clinical reasoning. By orchestrating specialized TxAgents (Executors) with long-context foundation models (Analysts), we mitigate the reasoning deficits observed in monolithic models. Beyond simple modularity, we demonstrate that a Stratified Ensemble strategy significantly outperforms global pooling by preserving evidentiary diversity, effectively addressing the information bottleneck. Furthermore, our stress tests reveal critical scaling insights: (1) a \textit{Context-Performance Paradox}, where extending reasoning contexts beyond 12k tokens introduces noise that degrades accuracy; and (2) the \textit{Curse of Dimensionality} in action spaces, where expanding toolsets necessitates hierarchical retrieval strategies. Crucially, our approach underscores the potential of training-free architectural engineering, achieving state-of-the-art performance on CURE-Bench without the need for expensive end-to-end finetuning. This provides a scalable, agile foundation for the next generation of trustworthy AI-driven therapeutics. Code has been released on https://github.com/June01/CureAgent.

cs.AI↗

arXiv2025

TokenScale: Timely and Accurate Autoscaling for Disaggregated LLM Serving with Token Velocity

The architectural shift to prefill/decode (PD) disaggregation in LLM serving improves resource utilization but struggles with the bursty nature of modern workloads. Existing autoscaling policies, often retrofitted from monolithic systems like those in AIBrix and DistServe, rely on lagging indicators such as GPU utilization or coarse-grained request counts. This results in slow reactions to load spikes, leading to significant Time-to First-Token (TTFT) and Time-Per-Output-Token (TPOT) SLO violations and costly over-provisioning. We introduce TokenScale, an autoscaling framework that resolves this performance mismatch through two innovations. First, we propose Token Velocity, a novel metric that unifies the prefill, network, and decode stages by quantifying their rate of work. As a leading indicator of system backpressure, it enables proactive scaling. Second, Convertible Decoders allow decoder GPUs to dynamically execute prefill tasks during traffic spikes, creating a rapid-response buffer that absorbs bursts and eliminates the initialization latency of new prefillers. Our evaluation on a GPU cluster with production traces shows TokenScale improves SLO attainment from 50-88% to 80-96% and reduces costs by 4-14% over state-of-the-art systems, including DistServe, BlitzScale, and AIBrix. By uniting a predictive metric with a flexible system design, TokenScale significantly boosts the performance and efficiency of disaggregated LLM serving infrastructure.

cs.DC↗

arXiv2025

Quantifying the Limits of Segmentation Foundation Models: Modeling Challenges in Segmenting Tree-Like and Low-Contrast Objects

Image segmentation foundation models (SFMs) like Segment Anything Model (SAM) have achieved impressive zero-shot and interactive segmentation across diverse domains. However, they struggle to segment objects with certain structures, particularly those with dense, tree-like morphology and low textural contrast from their surroundings. These failure modes are crucial for understanding the limitations of SFMs in real-world applications. To systematically study this issue, we introduce interpretable metrics quantifying object tree-likeness and textural separability. On carefully controlled synthetic experiments and real-world datasets, we show that SFM performance (\eg, SAM, SAM 2, HQ-SAM) noticeably correlates with these factors. We attribute these failures to SFMs misinterpreting local structure as global texture, resulting in over-segmentation or difficulty distinguishing objects from similar backgrounds. Notably, targeted fine-tuning fails to resolve this issue, indicating a fundamental limitation. Our study provides the first quantitative framework for modeling the behavior of SFMs on challenging structures, offering interpretable insights into their segmentation capabilities.

cs.CV↗

arXiv2025

MCOP: Multi-UAV Collaborative Occupancy Prediction

Unmanned Aerial Vehicle (UAV) swarm systems necessitate efficient collaborative perception mechanisms for diverse operational scenarios. Current Bird's Eye View (BEV)-based approaches exhibit two main limitations: bounding-box representations fail to capture complete semantic and geometric information of the scene, and their performance significantly degrades when encountering undefined or occluded objects. To address these limitations, we propose a novel multi-UAV collaborative occupancy prediction framework. Our framework effectively preserves 3D spatial structures and semantics through integrating a Spatial-Aware Feature Encoder and Cross-Agent Feature Integration. To enhance efficiency, we further introduce Altitude-Aware Feature Reduction to compactly represent scene information, along with a Dual-Mask Perceptual Guidance mechanism to adaptively select features and reduce communication overhead. Due to the absence of suitable benchmark datasets, we extend three datasets for evaluation: two virtual datasets (Air-to-Pred-Occ and UAV3D-Occ) and one real-world dataset (GauUScene-Occ). Experiments results demonstrate that our method achieves state-of-the-art accuracy, significantly outperforming existing collaborative methods while reducing communication overhead to only a fraction of previous approaches.

cs.CV↗

arXiv2025

Why and when merging surface nanobubbles jump

Gas bubble accumulation on substrates reduces the efficiency of many physicochemical processes, such as water electrolysis. For microbubbles, where buoyancy is negligible, coalescence-induced jumping driven by the release of surface energy provides an efficient pathway for their early detachment. At the nanoscale, however, gas compressibility breaks volume conservation during coalescence, suppressing surface energy release and seemingly disabling this detachment route. Using molecular dynamics simulations, continuum numerical simulations, and theoretical analysis, we show that surface nanobubbles with sufficiently large contact angles can nevertheless detach after coalescence. In this regime, detachment is powered by the release of pressure energy associated with nanobubble volume expansion. This finding thus establishes a unified driving mechanism for coalescence-induced bubble detachment across all length scales.

cond-mat.soft↗

arXiv2025

Joint Classification of Haze and Dust Events Using Factorial Hidden Markov Model Framework

Haze and dust pollution events have significant adverse impacts on human health and ecosystems. Their formation-impact interactions are complex, creating substantial modeling and computational challenges for joint classification. To address the state-space explosion faced by conventional Hidden Markov Models in multivariate dynamic settings, this study develops a classification framework based on the Factorial Hidden Markov Model. The framework assumes statistical independence across multiple latent chains and applies the Walsh-Hadamard transform to reduce computational and memory costs. A Gaussian copula decouples marginal distributions from dependence to capture nonlinear correlations among meteorological and pollution indicators. Algorithmically, mutual information weights the observational variables to increase the sensitivity of Viterbi decoding to salient features, and a single global weight hyperparameter balances emission and transition contributions in the decoding objective. In an empirical application, the model attains a Micro-F1 of 0.9459; for the low-frequency classes Dust prevalence below 1\% and Haze prevalence below 10\%, the F1-scores improve from 0.19 and 0.32 under a baseline FHMM to 0.75 and 0.68. The framework provides a scalable pathway for statistical modeling of complex air-pollution events and supplies quantitative evidence for decision-making in outdoor activity management and fine-grained environmental governance.

stat.AP↗

arXiv2025

Year-over-Year Developments in Financial Fraud Detection via Deep Learning: A Systematic Literature Review

This paper systematically reviews advancements in deep learning (DL) techniques for financial fraud detection, a critical issue in the financial sector. Using the Kitchenham systematic literature review approach, 57 studies published between 2019 and 2024 were analyzed. The review highlights the effectiveness of various deep learning models such as Convolutional Neural Networks, Long Short-Term Memory, and transformers across domains such as credit card transactions, insurance claims, and financial statement audits. Performance metrics such as precision, recall, F1-score, and AUC-ROC were evaluated. Key themes explored include the impact of data privacy frameworks and advancements in feature engineering and data preprocessing. The study emphasizes challenges such as imbalanced datasets, model interpretability, and ethical considerations, alongside opportunities for automation and privacy-preserving techniques such as blockchain integration and Principal Component Analysis. By examining trends over the past five years, this review identifies critical gaps and promising directions for advancing DL applications in financial fraud detection, offering actionable insights for researchers and practitioners.

cs.LG↗

arXiv2025

DAG-AFL:Directed Acyclic Graph-based Asynchronous Federated Learning

Due to the distributed nature of federated learning (FL), the vulnerability of the global model and the need for coordination among many client devices pose significant challenges. As a promising decentralized, scalable and secure solution, blockchain-based FL methods have attracted widespread attention in recent years. However, traditional consensus mechanisms designed for Proof of Work (PoW) similar to blockchain incur substantial resource consumption and compromise the efficiency of FL, particularly when participating devices are wireless and resource-limited. To address asynchronous client participation and data heterogeneity in FL, while limiting the additional resource overhead introduced by blockchain, we propose the Directed Acyclic Graph-based Asynchronous Federated Learning (DAG-AFL) framework. We develop a tip selection algorithm that considers temporal freshness, node reachability and model accuracy, with a DAG-based trusted verification strategy. Extensive experiments on 3 benchmarking datasets against eight state-of-the-art approaches demonstrate that DAG-AFL significantly improves training efficiency and model accuracy by 22.7% and 6.5% on average, respectively.

cs.LG↗

arXiv2025

Antiferromagnetism and Tightly Bound Cooper Pairs Induced by Kinetic Frustration

Antiferromagnetism and superconductivity are often viewed as competing orders in correlated electron systems. Here, we demonstrate that kinetic frustration in hole motion facilitates their coexistence within the square-lattice repulsive Hubbard model. Combining exact analytical solutions on tailored geometries with large-scale numerical simulations, we reveal a robust pairing mechanism: holes on opposite sublattices behave as if they carry opposite effective charges due to spin singlet formation from kinetic frustration. This emergent property suppresses phase separation and fosters a coherent $d$-wave superconducting channel embedded within a long-range antiferromagnetic background. Our findings establish a minimal yet broadly applicable framework for stabilizing strong-coupling superconductivity in doped Mott insulators.

cond-mat.str-el↗

arXiv2025

LLM Agent for Hyper-Parameter Optimization

Hyper-parameters are essential and critical for the performance of communication algorithms. However, current hyper-parameters optimization approaches for Warm-Start Particles Swarm Optimization with Crossover and Mutation (WS-PSO-CM) algorithm, designed for radio map-enabled unmanned aerial vehicle (UAV) trajectory and communication, are primarily heuristic-based, exhibiting low levels of automation and improvable performance. In this paper, we design an Large Language Model (LLM) agent for automatic hyper-parameters-tuning, where an iterative framework and Model Context Protocol (MCP) are applied. In particular, the LLM agent is first set up via a profile, which specifies the boundary of hyper-parameters, task objective, terminal condition, conservative or aggressive strategy of optimizing hyper-parameters, and LLM configurations. Then, the LLM agent iteratively invokes WS-PSO-CM algorithm for exploration. Finally, the LLM agent exits the loop based on the terminal condition and returns an optimized set of hyperparameters. Our experiment results show that the minimal sum-rate achieved by hyper-parameters generated via our LLM agent is significantly higher than those by both human heuristics and random generation methods. This indicates that an LLM agent with PSO and WS-PSO-CM algorithm knowledge is useful in seeking high-performance hyper-parameters.

cs.IT↗

arXiv2025

Cooperative Sensing in Cell-free Massive MIMO ISAC Systems: Performance Optimization and Signal Processing

Integrated sensing and communication (ISAC), as a technology enabled seamless connection between communication and sensing, is regarded a core enabling technology for these applications. However, the accuracy of single-node sensing in ISAC system is limited, prompting the emergence of multi-node cooperative sensing. In multi-node cooperative sensing, the synchronization error limits the sensing accuracy, which can be mitigated by the architecture of cell-free massive multi-input multi-output (CF-mMIMO), since the multiple nodes are interconnected via optical fibers with high synchronization accuracy. However, the multi-node cooperative sensing in CF-mMIMO ISAC systems faces the following challenges: 1) The joint optimization of placement and resource allocation of distributed access points (APs) to improve the sensing performance in multi-target detection scenario is difficult; 2) The fusion of the sensing information from distributed APs with multi-view discrepancies is difficult. To address these challenges, this paper proposes a joint placement and antenna resource optimization scheme for distributed APs to minimize the sensing Cramr-Rao bound for targets' parameters within the area of interest. Then, a symbol-level fusion-based multi-dynamic target sensing (SL-MDTS) scheme is provided, effectively fusing sensing information from multiple APs. The simulation results validate the effectiveness of the joint optimization scheme and the superiority of the SL-MDTS scheme. Compared to state-of-the-art grid-based symbol-level sensing information fusion schemes, the proposed SL-MDTS scheme improves the accuracy of localization and velocity estimation by 44 % and 41.4 %, respectively.

eess.SP↗

arXiv2025

I Know What You Said: Unveiling Hardware Cache Side-Channels in Local Large Language Model Inference

Large Language Models (LLMs) that can be deployed locally have recently gained popularity for privacy-sensitive tasks, with companies such as Meta, Google, and Intel playing significant roles in their development. However, the security of local LLMs through the lens of hardware cache side-channels remains unexplored. In this paper, we unveil novel side-channel vulnerabilities in local LLM inference: token value and token position leakage, which can expose both the victim's input and output text, thereby compromising user privacy. Specifically, we found that adversaries can infer the token values from the cache access patterns of the token embedding operation, and deduce the token positions from the timing of autoregressive decoding phases. To demonstrate the potential of these leaks, we design a novel eavesdropping attack framework targeting both open-source and proprietary LLM inference systems. The attack framework does not directly interact with the victim's LLM and can be executed without privilege. We evaluate the attack on a range of practical local LLM deployments (e.g., Llama, Falcon, and Gemma), and the results show that our attack achieves promising accuracy. The restored output and input text have an average edit distance of 5.2% and 17.3% to the ground truth, respectively. Furthermore, the reconstructed texts achieve average cosine similarity scores of 98.7% (input) and 98.0% (output).

cs.CR↗

arXiv2025

Target Semantics Clustering via Text Representations for Robust Universal Domain Adaptation

Universal Domain Adaptation (UniDA) focuses on transferring source domain knowledge to the target domain under both domain shift and unknown category shift. Its main challenge lies in identifying common class samples and aligning them. Current methods typically obtain target domain semantics centers from an unconstrained continuous image representation space. Due to domain shift and the unknown number of clusters, these centers often result in complex and less robust alignment algorithm. In this paper, based on vision-language models, we search for semantic centers in a semantically meaningful and discrete text representation space. The constrained space ensures almost no domain bias and appropriate semantic granularity for these centers, enabling a simple and robust adaptation algorithm. Specifically, we propose TArget Semantics Clustering (TASC) via Text Representations, which leverages information maximization as a unified objective and involves two stages. First, with the frozen encoders, a greedy search-based framework is used to search for an optimal set of text embeddings to represent target semantics. Second, with the search results fixed, encoders are refined based on gradient descent, simultaneously achieving robust domain alignment and private class clustering. Additionally, we propose Universal Maximum Similarity (UniMS), a scoring function tailored for detecting open-set samples in UniDA. Experimentally, we evaluate the universality of UniDA algorithms under four category shift scenarios. Extensive experiments on four benchmarks demonstrate the effectiveness and robustness of our method, which has achieved state-of-the-art performance.

cs.CV↗

arXiv2025

Evidence of competing ground states between fractional Chern insulator and antiferromagnetism in moiré MoTe2

Two-dimensional moire materials present unprecedented opportunities to explore quantum phases of matter arising from the interplay of band topology and strong correlations.One of the most striking examples is the recent observation of fractional quantum anomalous Hall (FQAH) effect in twisted bilayer MoTe$_2$ (tMoTe2) with relatively large twist angles(~3.7deg-3.9deg). The electronic ground states are usually expected to be sensitive to the twist angle, as the twist angle determines the electron bandwidth and correlation strength in the moire system. Here, we report the observation of unexpected competing magnetic ground states in tMoTe2 moire superlattice, on which balance can be tipped by both twist angle and electric field (E). Specifically, we observed anomalous antiferromagnetic (AFM) ground states with zero Hall resistance at both v_h=1 and 2/3, at intermediate twist angles ~3deg. The AFM orders are suppressed by applying vertical E, and emergent ferromagnetism accompanied by integer Chern insulator (ICI) or fractional Chern insulator (FCI) states are observed near the critical E (E_c) of moire superlattice symmetry transition. Our results demonstrate tMoTe2 as a fascinating platform for exploring unexpected correlated phases with nontrivial topology and fractional excitations and point to electric-field-controlled ultralow-power spin-valleytronic devices.

cond-mat.mes-hall↗

arXiv2025

The Role of Machine Learning in Reducing Healthcare Costs: The Impact of Medication Adherence and Preventive Care on Hospitalization Expenses

This study reveals the important role of prevention care and medication adherence in reducing hospitalizations. By using a structured dataset of 1,171 patients, four machine learning models Logistic Regression, Gradient Boosting, Random Forest, and Artificial Neural Networks are applied to predict five-year hospitalization risk, with the Gradient Boosting model achieving the highest accuracy of 81.2%. The result demonstrated that patients with high medication adherence and consistent preventive care can reduce 38.3% and 37.7% in hospitalization risk. The finding also suggests that targeted preventive care can have positive Return on Investment (ROI), and therefore ML models can effectively direct personalized interventions and contribute to long-term medical savings.

cs.LG↗

arXiv2025

LVC: A Lightweight Compression Framework for Enhancing VLMs in Long Video Understanding

Long video understanding is a complex task that requires both spatial detail and temporal awareness. While Vision-Language Models (VLMs) obtain frame-level understanding capabilities through multi-frame input, they suffer from information loss due to the sparse sampling strategy. In contrast, Video Large Language Models (Video-LLMs) capture temporal relationships within visual features but are limited by the scarcity of high-quality video-text datasets. To transfer long video understanding capabilities to VLMs with minimal data and computational cost, we propose Lightweight Video Compression (LVC), a novel method featuring the Query-Attention Video Compression mechanism, which effectively tackles the sparse sampling problem in VLMs. By training only the alignment layer with 10k short video-text pairs, LVC significantly enhances the temporal reasoning abilities of VLMs. Extensive experiments show that LVC provides consistent performance improvements across various models, including the InternVL2 series and Phi-3.5-Vision. Notably, the InternVL2-40B-LVC achieves scores of 68.2 and 65.9 on the long video understanding benchmarks MLVU and Video-MME, respectively, with relative improvements of 14.6% and 7.7%. The enhanced models and code will be publicly available soon.

cs.CV↗