arXiv subjects

Xuan Wang

Publications and source records attributed to Xuan Wang.

At least 109 records · Page 6Linked to original sources

arXiv2025

AIMNET: An IoT-Empowered Digital Twin for Continuous Gas Emission Monitoring and Early Hazard Detection

A Digital Twin (DT) framework to enhance carbon-based gas plume monitoring is critical for supporting timely and effective mitigation responses to environmental hazards such as industrial gas leaks, or wildfire outbreaks carrying large carbon emissions. We present AIMNET, a one-of-a-kind DT framework that integrates a built-in-house Internet of Things (IoT)-based continuous sensing network with a physics-based multi-scale weather-gas transport model, that enables high-resolution and real-time simulation and detection of carbon gas emissions. AIMNET features a three-layer system architecture: (i) physical world: custom-built devices for continuous monitoring; (ii) bidirectional information feedback links: intelligent data transmission and reverse control; and (iii) digital twin world: AI-driven analytics for prediction, anomaly detection, and dynamic weather-gas coupled molecule transport modeling. Designed for scalable, energy-efficient deployment in remote environments, AIMNET architecture is realized through a small-scale distributed sensing network over an oil and gas production basin. To demonstrate the high-resolution, fast-responding concept, an equivalent mobile-based emission monitoring network was deployed around a wastewater treatment plant that constantly emits methane plumes. Our preliminary results through which, have successfully captured the methane emission events whose dynamics have been further resolved by the tiered model simulations. This work supports our position that AIMNET provides a promising DT framework for reliable, real-time monitoring and predictive risk assessment. In the end, we also discuss key implementation challenges and outline future directions for advancing such a new DT framework for translation deployment.

cs.NI↗

arXiv2025

Scaling Agentic Reinforcement Learning for Tool-Integrated Reasoning in VLMs

While recent vision-language models (VLMs) demonstrate strong image understanding, their ability to "think with images", i.e., to reason through multi-step visual interactions, remains limited. We introduce VISTA-Gym, a scalable training environment for incentivizing tool-integrated visual reasoning capabilities in VLMs. VISTA-Gym unifies diverse real-world multimodal reasoning tasks (7 tasks from 13 datasets in total) with a standardized interface for visual tools (e.g., grounding, parsing), executable interaction loops, verifiable feedback signals, and efficient trajectory logging, enabling visual agentic reinforcement learning at scale. While recent VLMs exhibit strong text-only reasoning, both proprietary and open-source models still struggle with tool selection, invocation, and coordination. With VISTA-Gym, we train VISTA-R1 to interleave tool-use with agentic reasoning via multi-turn trajectory sampling and end-to-end reinforcement learning. Extensive experiments across 11 public reasoning-intensive VQA benchmarks show that VISTA-R1-8B outperforms state-of-the-art baselines with similar sizes by 9.51%-18.72%, demonstrating VISTA-Gym as an effective training ground to unlock the tool-integrated reasoning capabilities for VLMs.

cs.AI↗

arXiv2025

LLM4Cell: A Survey of Large Language and Agentic Models for Single-Cell Biology

Large language models (LLMs) and emerging agentic frameworks are beginning to transform single-cell biology by enabling natural-language reasoning, generative annotation, and multimodal data integration. However, progress remains fragmented across data modalities, architectures, and evaluation standards. LLM4Cell presents the first unified survey of 58 foundation and agentic models developed for single-cell research, spanning RNA, ATAC, multi-omic, and spatial modalities. We categorize these methods into five families-foundation, text-bridge, spatial, multimodal, epigenomic, and agentic-and map them to eight key analytical tasks including annotation, trajectory and perturbation modeling, and drug-response prediction. Drawing on over 40 public datasets, we analyze benchmark suitability, data diversity, and ethical or scalability constraints, and evaluate models across 10 domain dimensions covering biological grounding, multi-omics alignment, fairness, privacy, and explainability. By linking datasets, models, and evaluation domains, LLM4Cell provides the first integrated view of language-driven single-cell intelligence and outlines open challenges in interpretability, standardization, and trustworthy model development.

cs.CL↗

arXiv2025

JudgeBoard: Benchmarking and Enhancing Small Language Models for Reasoning Evaluation

While small language models (SLMs) have shown promise on various reasoning tasks, their ability to judge the correctness of answers remains unclear compared to large language models (LLMs). Prior work on LLM-as-a-judge frameworks typically relies on comparing candidate answers against ground-truth labels or other candidate answers using predefined metrics like entailment. However, this approach is inherently indirect and difficult to fully automate, offering limited support for fine-grained and scalable evaluation of reasoning outputs. In this work, we propose JudgeBoard, a novel evaluation pipeline that directly queries models to assess the correctness of candidate answers without requiring extra answer comparisons. We focus on two core reasoning domains: mathematical reasoning and science/commonsense reasoning, and construct task-specific evaluation leaderboards using both accuracy-based ranking and an Elo-based rating system across five benchmark datasets, enabling consistent model comparison as judges rather than comparators. To improve judgment performance in lightweight models, we propose MAJ (Multi-Agent Judging), a novel multi-agent evaluation framework that leverages multiple interacting SLMs with distinct reasoning profiles to approximate LLM-level judgment accuracy through collaborative deliberation. Experimental results reveal a significant performance gap between SLMs and LLMs in isolated judging tasks. However, our MAJ framework substantially improves the reliability and consistency of SLMs. On the MATH dataset, MAJ using smaller-sized models as backbones performs comparatively well or even better than their larger-sized counterparts. Our findings highlight that multi-agent SLM systems can potentially match or exceed LLM performance in judgment tasks, with implications for scalable and efficient assessment.

cs.AI↗

arXiv2025

X-MoGen: Unified Motion Generation across Humans and Animals

Text-driven motion generation has attracted increasing attention due to its broad applications in virtual reality, animation, and robotics. While existing methods typically model human and animal motion separately, a joint cross-species approach offers key advantages, such as a unified representation and improved generalization. However, morphological differences across species remain a key challenge, often compromising motion plausibility. To address this, we propose X-MoGen, the first unified framework for cross-species text-driven motion generation covering both humans and animals. X-MoGen adopts a two-stage architecture. First, a conditional graph variational autoencoder learns canonical T-pose priors, while an autoencoder encodes motion into a shared latent space regularized by morphological loss. In the second stage, we perform masked motion modeling to generate motion embeddings conditioned on textual descriptions. During training, a morphological consistency module is employed to promote skeletal plausibility across species. To support unified modeling, we construct UniMo4D, a large-scale dataset of 115 species and 119k motion sequences, which integrates human and animal motions under a shared skeletal topology for joint training. Extensive experiments on UniMo4D demonstrate that X-MoGen outperforms state-of-the-art methods on both seen and unseen species.

cs.CV↗

arXiv2025

An exploration of the noise sensitivity of the Shor's algorithm

Quantum algorithms face significant challenges due to qubit susceptibility to environmental noise, and quantum error correction typically requires prohibitive resource overhead. This paper proposes that quantum algorithms may possess inherent noise resilience characteristics that could reduce implementation barriers. We investigate Shor's algorithm by applying circuit-level noise models directly to the original algorithm circuit. Our findings reveal that Shor's algorithm demonstrates superior fault tolerance under Z noise compared to X and Y noise. Focusing on the modular exponentiation circuit which is the core component of the algorithm, we conduct fault-tolerant position statistics on circuits with bit lengths from 4 to 9. The results show that under Z noise, fault-tolerant positions grow with the same quartic polynomial order as potential error positions as the problem scale increases. In contrast, fault tolerance under X and Y noise exhibits a strong dependence on the composite number N and the parameter a. Based on these findings, we develop an extrapolation method predicting that the minimum probability of a correct output of the modular exponentiation circuit to factor 2048 bit integers under biased noise is approximately 1.417*{10}^{-17}.

quant-ph↗

arXiv2025

Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs

Recent advances in Multimodal Large Language Models (MLLMs) have enhanced their versatility as they integrate a growing number of modalities. Considering the heavy cost of training MLLMs, it is efficient to reuse the existing ones and extend them to more modalities through Modality-incremental Continual Learning (MCL). The exploration of MCL is in its early stages. In this work, we dive into the causes of performance degradation in MCL. We uncover that it suffers not only from forgetting as in traditional continual learning, but also from misalignment between the modality-agnostic and modality-specific components. To this end, we propose an elegantly simple MCL paradigm called "MErge then ReAlign" (MERA) to address both forgetting and misalignment. MERA avoids introducing heavy model budgets or modifying model architectures, hence is easy to deploy and highly reusable in the MLLM community. Extensive experiments demonstrate the impressive performance of MERA, holding an average of 99.84\% Backward Relative Gain when extending to four modalities, achieving nearly lossless MCL performance. Our findings underscore the misalignment issue in MCL. More broadly, our work showcases how to adjust different components of MLLMs during continual learning.

cs.LG↗

arXiv2025

A Wideband Tunable, Nonreciprocal Bandpass Filter Using Magnetostatic Surface Waves with Zero Static Power Consumption

Modern wireless systems demand compact, power-efficient RF front-end components that support wideband tunability and nonreciprocity. We present a new class of miniature bandpass filter that achieves both continuously tunable frequency operation (4-17.7 GHz) and high nonreciprocity (>25 dB), all within a compact size of 1.07 cm3. The filter employs a microfabricated 18 micrometer thick Yttrium Iron Garnet waveguide with meander-line aluminum transducers, enabling low-loss unidirectional propagation via magnetostatic surface waves. Leveraging a benzocyclobutene planarization fabrication process, this study enables a dispersion profile unique to thick YIG films, resulting in enhanced filter skirt performance with minimal spurious modes. Frequency tuning is enabled by a zero-static-power magnetic bias circuit using transient current pulses, eliminating continuous power consumption. The filter demonstrates low insertion loss (3-5 dB), high out-of-band rejection (>30 dB), narrow bandwidth (100-200 MHz), robust power handling (>10.4 dBm), and high linearity (IIP3 > 26 dBm).

physics.app-ph↗

arXiv2025

OPTAGENT: Optimizing Multi-Agent LLM Interactions Through Verbal Reinforcement Learning for Enhanced Reasoning

Large Language Models (LLMs) have shown remarkable reasoning capabilities in mathematical and scientific tasks. To enhance complex reasoning, multi-agent systems have been proposed to harness the collective intelligence of LLM agents. However, existing collaboration structures are either predefined or rely on majority voting or round-table debates, which can suppress correct but less dominant agent contributions. Recent approaches model multi-agent systems as graph networks but optimize purely for agent performance, neglecting the quality of interactions. We hypothesize that effective agent communication is crucial for multi-agent reasoning and that debating quality plays a significant role. To address this, we propose $\ours$, a multi-agent verbal reinforcement learning algorithm that dynamically constructs and refines multi-agent collaboration structures. Our method defines action spaces and a feedback mechanism that evaluates communication robustness and coherence throughout the debate. The final decision is achieved through a majority vote over all the agents. We assess $\ours$ on various reasoning tasks, including mathematical reasoning, creative writing, scientific reasoning, and numerical sorting. Results demonstrate that our approach significantly outperforms single-agent prompting methods and state-of-the-art multi-agent frameworks on diverse tasks.

cs.AI↗

arXiv2025

Sequential Comics for Jailbreaking Multimodal Large Language Models via Structured Visual Storytelling

Multimodal large language models (MLLMs) exhibit remarkable capabilities but remain susceptible to jailbreak attacks exploiting cross-modal vulnerabilities. In this work, we introduce a novel method that leverages sequential comic-style visual narratives to circumvent safety alignments in state-of-the-art MLLMs. Our method decomposes malicious queries into visually innocuous storytelling elements using an auxiliary LLM, generates corresponding image sequences through diffusion models, and exploits the models' reliance on narrative coherence to elicit harmful outputs. Extensive experiments on harmful textual queries from established safety benchmarks show that our approach achieves an average attack success rate of 83.5\%, surpassing prior state-of-the-art by 46\%. Compared with existing visual jailbreak methods, our sequential narrative strategy demonstrates superior effectiveness across diverse categories of harmful content. We further analyze attack patterns, uncover key vulnerability factors in multimodal safety mechanisms, and evaluate the limitations of current defense strategies against narrative-driven attacks, revealing significant gaps in existing protections.

cs.CR↗

arXiv2025

Agentic Reasoning and Refinement through Semantic Interaction

Sensemaking report writing often requires multiple refinements in the iterative process. While Large Language Models (LLMs) have shown promise in generating initial reports based on human visual workspace representations, they struggle to precisely incorporate sequential semantic interactions during the refinement process. We introduce VIS-ReAct, a framework that reasons about newly-added semantic interactions in visual workspaces to steer the LLM for report refinement. VIS-ReAct is a two-agent framework: a primary LLM analysis agent interprets new semantic interactions to infer user intentions and generate refinement planning, followed by an LLM refinement agent that updates reports accordingly. Through case study, VIS-ReAct outperforms baseline and VIS-ReAct (without LLM analysis) on targeted refinement, semantic fidelity, and transparent inference. Results demonstrate that VIS-ReAct better handles various interaction types and granularities while enhancing the transparency of human-LLM collaboration.

cs.HC↗

arXiv2025

Data-Driven Stabilization of Unknown Linear-Threshold Network Dynamics

This paper studies the data-driven control of unknown linear-threshold network dynamics to stabilize the state to a reference value. We consider two types of controllers: (i) a state feedback controller with feed-forward reference input and (ii) an augmented feedback controller with error integration. The first controller features a simpler structure and is easier to design, while the second offers improved performance in the presence of system parameter changes and disturbances. Our design strategy employs state-input datasets to construct data-based representations of the closed-loop dynamics. Since these representations involve linear threshold functions, we rewrite them as switched linear systems, and formulate the design problem as that of finding a common controller for all the resulting modes. This gives rise to a set of linear matrix inequalities (LMIs) whose solutions corresponds to the controller gain matrices. We analyze the computational complexity of solving the LMIs and propose a simplified, sufficient set of conditions that scales linearly with the system state. Simulations on two case studies involving regulation of firing rate dynamics in rodent brains and of arousal level dynamics in humans demonstrate the effectiveness of the controller designs.

eess.SY↗

arXiv2025

Reconstructing Brain Causal Dynamics for Subject and Task Fingerprints using fMRI Time-series Data

Purpose: Recently, there has been a revived interest in system neuroscience causation models, driven by their unique capability to unravel complex relationships in multi-scale brain networks. In this paper, we present a novel method that leverages causal dynamics to achieve effective fMRI-based subject and task fingerprinting. Methods: By applying an implicit-explicit discretization scheme, we develop a two-timescale linear state-space model. Through data-driven identification of its parameters, the model captures causal signatures, including directed interactions among brain regions from a spatial perspective, and disentangled fast and slow dynamic modes of brain activity from a temporal perspective. These causal signatures are then integrated with: (i) a modal decomposition and projection method for model-based subject identification, and (ii) a Graph Neural Network (GNN) framework for learning-based task classification. Furthermore, we introduce the concept of the brain reachability landscape as a novel visualization tool, which quantitatively characterizes the maximum possible activation levels of brain regions under various fMRI tasks. Results: We evaluate the proposed approach using the Human Connectome Project dataset and demonstrate its advantage over non-causality-based methods. The obtained causal signatures are visualized and demonstrate clear biological relevance with established understandings of brain function. Conclusion: We verified the feasibility and effectiveness of utilizing brain causal signatures for subject and task fingerprinting. Additionally, our work paves the way for further studies on causal fingerprints with potential applications in both healthy controls and neurodegenerative diseases.

eess.SY↗

arXiv2025

Towards Reasoning Ability of Small Language Models

Reasoning has long been viewed as an emergent property of large language models (LLMs). However, recent studies challenge this assumption, showing that small language models (SLMs) can also achieve competitive reasoning performance. This paper introduces ThinkSLM, the first extensive benchmark to systematically evaluate and study the reasoning abilities of SLMs trained from scratch or derived from LLMs through quantization, pruning, and distillation. We first establish a reliable evaluation criterion comparing available methods and LLM judges against our human evaluations. Then we present a study evaluating 72 diverse SLMs from six major model families across 17 reasoning benchmarks. We repeat all our experiments three times to ensure a robust assessment. Our findings show that: 1) reasoning ability in SLMs is strongly influenced by training methods and data quality rather than solely model scale; 2) quantization preserves reasoning capability, while pruning significantly disrupts it; 3) larger models consistently exhibit higher robustness against adversarial perturbations and intermediate reasoning, but certain smaller models closely match or exceed the larger models' performance. Our findings challenge the assumption that scaling is the only way to achieve strong reasoning. Instead, we foresee a future where SLMs with strong reasoning capabilities can be developed through structured training or post-training compression. Our ThinkSLM Leaderboard is publicly available at: https://ctrl-gaurav.github.io/thinkslm.github.io/

cs.CL↗

arXiv2025

DEBATE, TRAIN, EVOLVE: Self Evolution of Language Model Reasoning

Large language models (LLMs) have improved significantly in their reasoning through extensive training on massive datasets. However, relying solely on additional data for improvement is becoming increasingly impractical, highlighting the need for models to autonomously enhance their reasoning without external supervision. In this paper, we propose Debate, Train, Evolve (DTE), a novel ground truth-free training framework that uses multi-agent debate traces to evolve a single language model. We also introduce a new prompting strategy Reflect-Critique-Refine, to improve debate quality by explicitly instructing agents to critique and refine their reasoning. Extensive evaluations on seven reasoning benchmarks with six open-weight models show that our DTE framework achieve substantial improvements, with an average accuracy gain of 8.92% on the challenging GSM-PLUS dataset. Furthermore, we observe strong cross-domain generalization, with an average accuracy gain of 5.8% on all other benchmarks, suggesting that our method captures general reasoning capabilities. Our framework code and trained models are publicly available at https://github.com/ctrl-gaurav/Debate-Train-Evolve

cs.CL↗

arXiv2025

A Comprehensive Survey on the Trustworthiness of Large Language Models in Healthcare

The application of large language models (LLMs) in healthcare holds significant promise for enhancing clinical decision-making, medical research, and patient care. However, their integration into real-world clinical settings raises critical concerns around trustworthiness, particularly around dimensions of truthfulness, privacy, safety, robustness, fairness, and explainability. These dimensions are essential for ensuring that LLMs generate reliable, unbiased, and ethically sound outputs. While researchers have recently begun developing benchmarks and evaluation frameworks to assess LLM trustworthiness, the trustworthiness of LLMs in healthcare remains underexplored, lacking a systematic review that provides a comprehensive understanding and future insights. This survey addresses that gap by providing a comprehensive review of current methodologies and solutions aimed at mitigating risks across key trust dimensions. We analyze how each dimension affects the reliability and ethical deployment of healthcare LLMs, synthesize ongoing research efforts, and identify critical gaps in existing approaches. We also identify emerging challenges posed by evolving paradigms, such as multi-agent collaboration, multi-modal reasoning, and the development of small open-source medical models. Our goal is to guide future research toward more trustworthy, transparent, and clinically viable LLMs.

cs.CY↗

arXiv2025

Efficient Last-Iterate Convergence in Regret Minimization via Adaptive Reward Transformation

Regret minimization is a powerful method for finding Nash equilibria in Normal-Form Games (NFGs) and Extensive-Form Games (EFGs), but it typically guarantees convergence only for the average strategy. However, computing the average strategy requires significant computational resources or introduces additional errors, limiting its practical applicability. The Reward Transformation (RT) framework was introduced to regret minimization to achieve last-iterate convergence through reward function regularization. However, it faces practical challenges: its performance is highly sensitive to manually tuned parameters, which often deviate from theoretical convergence conditions, leading to slow convergence, oscillations, or stagnation in local optima. Inspired by previous work, we propose an adaptive technique to address these issues, ensuring better consistency between theoretical guarantees and practical performance for RT Regret Matching (RTRM), RT Counterfactual Regret Minimization (RTCFR), and their variants in solving NFGs and EFGs more effectively. Our adaptive methods dynamically adjust parameters, balancing exploration and exploitation while improving regret accumulation, ultimately enhancing asymptotic last-iterate convergence and achieving linear convergence. Experimental results demonstrate that our methods significantly accelerate convergence, outperforming state-of-the-art algorithms.

cs.GT↗

arXiv2025

QUADOL: A Quality-Driven Approximate Logic Synthesis Method Exploiting Dual-Output LUTs for Modern FPGAs

Approximate computing is a new computing paradigm. One important area of it is designing approximate circuits for FPGA. Modern FPGAs support dual-output LUT, which can significantly reduce the area of FPGA designs. Several existing works explored the use of dual-output in approximate computing. However, they are limited to small-scale arithmetic circuits. To address the problem, this work proposes QUADOL, a quality-driven ALS method by exploiting dual-output LUTs for modern FPGAs. We propose a technique to approximately merge two single-output LUTs (i.e., a LUT pair) into a dual-output LUT. In addition, we transform the problem of selecting multiple LUT pairs for simultaneous approximate merging into a maximum matching problem to maximize the area reduction. Since QUADOL exploits a new dimension, i.e., approximately merging a LUT pair into a dual-output LUT, it can be integrated with any existing ALS methods to strengthen them. Therefore, we also introduce QUADOL+, which is a generic framework to integrate QUADOL into existing ALS methods. The experimental results showed that QUADOL+ can reduce the LUT count by up to 18% compared to the state-of-the-art ALS methods for FPGA. Moreover, the approximate multipliers optimized by QUADOL+ dominate most prior FPGA-based approximate multipliers in the area-error plane.

cs.AR↗