arXiv subjects

Shuo Zhang

Publications and source records attributed to Shuo Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2025

BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping

Reinforcement learning (RL) has recently become the core paradigm for aligning and strengthening large language models (LLMs). Yet, applying RL in off-policy settings--where stale data from past policies are used for training--improves sample efficiency, but remains challenging: policy entropy declines sharply, optimization often becomes unstable and may even collapse. Through theoretical and empirical analysis, we identify two key insights: (i) an imbalance in optimization, where negative-advantage samples dominate the policy gradient, suppressing useful behaviors and risking gradient explosions; and (ii) the derived Entropy-Clip Rule, which reveals that the fixed clipping mechanism in PPO-like objectives systematically blocks entropy-increasing updates, thereby driving the policy toward over-exploitation at the expense of exploration. Building on these insights, we propose BAlanced Policy Optimization with Adaptive Clipping (BAPO), a simple yet effective method that dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization. Across diverse off-policy scenarios--including sample replay and partial rollout--BAPO achieves fast, stable, and data-efficient training. On AIME 2024 and AIME 2025 benchmarks, our 7B BAPO model surpasses open-source counterparts such as SkyWork-OR1-7B, while our 32B BAPO model not only achieves state-of-the-art results among models of the same scale but also outperforms leading proprietary systems like o3-mini and Gemini-2.5-Flash-Thinking.

cs.LG↗

arXiv2025

Nonconforming finite element spaces for $HΛ^k$ in $\mathbb{R}^n$

This paper constructs a unified family of nonconforming finite element spaces for $HΛ^k$ in $\mathbb{R}^n$ ($0\leqslant k\leqslant n$, $n\geqslant 1$). The spaces employ piecewise Whitney forms as shape functions, and include the lowest-degree Crouzeix-Raviart element space for $HΛ^0$. Optimal approximations and uniform discrete Poincaré inequalities are presented. Based on the newly constructed finite element spaces, discrete de Rham complexes with commutative diagrams, and the discrete Helmholtz decomposition and Hodge decomposition for piecewise constant spaces are established. All discrete operators involved are local, acting cell-wise. A framework of nonconforming finite element exterior calculus is then established, and is naturally connected to the classical conforming one. The cooperation of conforming and nonconforming finite element spaces leads to new discretization schemes of the Hodge Laplace problem. The new finite element spaces are constructed by a novel approach that seeks to mimic the dual connections between adjoint operators; novel construction methods and basic estimations are presented. Although the new spaces do not fit Ciarlet's finite element definition, they admit locally supported basis functions each spanning at most two adjacent cells, which makes the computation of the local stiffness matrices and the assembling of the global stiffness matrix implementable by following the standard procedure. Some numerical experiments are given to show the implementability and the performance of the new kind of spaces.

math.NA↗

arXiv2025

Qwen3Guard Technical Report

As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary "safe/unsafe" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with streaming LLM inference, thereby preventing timely intervention during generation and increasing exposure to harmful partial outputs. To address these challenges, we present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants: Generative Qwen3Guard, which casts safety classification as an instruction-following task to enable fine-grained tri-class judgments (safe, controversial, unsafe); and Stream Qwen3Guard, which introduces a token-level classification head for real-time safety monitoring during incremental text generation. Both variants are available in three sizes (0.6B, 4B, and 8B parameters) and support up to 119 languages and dialects, providing comprehensive, scalable, and low-latency safety moderation for global LLM deployments. Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification. All models are released under the Apache 2.0 license for public use.

cs.CL↗

arXiv2025

Precision measurement and modelling of the threshold-free 210Pb β spectrum

Beta decay is a fundamental process that governs nuclear stability and serves as a sensitive probe of the weak interaction and possible physics beyond the Standard Model of particle physics. However, precise measurements of complete $β$ decay spectra, particularly at low energies, remain experimentally and theoretically challenging. Here we report a high-precision, threshold-free measurement of the full $β$ decay spectrum of 210Pb to excited states of 210Bi, using a transition-edge sensor (TES)-based micro-calorimeter. This approach enables the detection of $β$ particle energies from 0 keV up to their endpoint by coincidence summing with subsequent de-excitation energy, thereby eliminating reconstruction artifacts near zero energy that have traditionally limited low-energy spectral accuracy. To our knowledge, this is the first complete, high-precision $β$ decay spectrum from 0 keV. The data resolve theoretical uncertainties associated with the atomic quantum exchange (AQE) effect. An accompanying ab initio theoretical framework, incorporating atomic, leptonic, and nuclear components, predicts a statistically significant (7.2 {$σ$}) enhancement in $β$ emission probability near zero energy, in agreement with the measurement and in contrast to models that omit AQE corrections. These results provide a new benchmark for $β$ decay theory at low energies, deepen our understanding of the weak interaction, and establish a critical foundation for searches for new physics, including dark matter interactions and precision studies of neutrinos.

nucl-ex↗

arXiv2025

REMA: A Unified Reasoning Manifold Framework for Interpreting Large Language Model

Understanding how Large Language Models (LLMs) perform complex reasoning and their failure mechanisms is a challenge in interpretability research. To provide a measurable geometric analysis perspective, we define the concept of the Reasoning Manifold, a latent low-dimensional geometric structure formed by the internal representations corresponding to all correctly reasoned generations. This structure can be conceptualized as the embodiment of the effective thinking paths that the model has learned to successfully solve a given task. Based on this concept, we build REMA, a framework that explains the origins of failures by quantitatively comparing the spatial relationships of internal model representations corresponding to both erroneous and correct reasoning samples. Specifically, REMA first quantifies the geometric deviation of each erroneous representation by calculating its k-nearest neighbors distance to the approximated manifold formed by correct representations, thereby providing a unified failure signal. It then localizes the divergence points where these deviations first become significant by tracking this deviation metric across the model's layers and comparing it against a baseline of internal fluctuations from correct representations, thus identifying where the reasoning chain begins to go off-track. Our extensive experiments on diverse language and multimodal models and tasks demonstrate the low-dimensional nature of the reasoning manifold and the high separability between erroneous and correct reasoning representations. The results also validate the effectiveness of the REMA framework in analyzing the origins of reasoning failures. This research connects abstract reasoning failures to measurable geometric deviations in representations, providing new avenues for in-depth understanding and diagnosis of the internal computational processes of black-box models.

cs.AI↗

arXiv2025

Real-Time System for Audio-Visual Target Speech Enhancement

We present a live demonstration for RAVEN, a real-time audio-visual speech enhancement system designed to run entirely on a CPU. In single-channel, audio-only settings, speech enhancement is traditionally approached as the task of extracting clean speech from environmental noise. More recent work has explored the use of visual cues, such as lip movements, to improve robustness, particularly in the presence of interfering speakers. However, to our knowledge, no prior work has demonstrated an interactive system for real-time audio-visual speech enhancement operating on CPU hardware. RAVEN fills this gap by using pretrained visual embeddings from an audio-visual speech recognition model to encode lip movement information. The system generalizes across environmental noise, interfering speakers, transient sounds, and even singing voices. In this demonstration, attendees will be able to experience live audio-visual target speech enhancement using a microphone and webcam setup, with clean speech playback through headphones.

eess.AS↗

arXiv2025

Enhancing RAG Efficiency with Adaptive Context Compression

Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge but incurs significant inference costs due to lengthy retrieved contexts. While context compression mitigates this issue, existing methods apply fixed compression rates, over-compressing simple queries or under-compressing complex ones. We propose Adaptive Context Compression for RAG (ACC-RAG), a framework that dynamically adjusts compression rates based on input complexity, optimizing inference efficiency without sacrificing accuracy. ACC-RAG combines a hierarchical compressor (for multi-granular embeddings) with a context selector to retain minimal sufficient information, akin to human skimming. Evaluated on Wikipedia and five QA datasets, ACC-RAG outperforms fixed-rate methods and matches/unlocks over 4 times faster inference versus standard RAG while maintaining or improving accuracy.

cs.CL↗

arXiv2025

GitTaskBench: A Benchmark for Code Agents Solving Real-World Tasks Through Code Repository Leveraging

Beyond scratch coding, exploiting large-scale code repositories (e.g., GitHub) for practical tasks is vital in real-world software development, yet current benchmarks rarely evaluate code agents in such authentic, workflow-driven scenarios. To bridge this gap, we introduce GitTaskBench, a benchmark designed to systematically assess this capability via 54 realistic tasks across 7 modalities and 7 domains. Each task pairs a relevant repository with an automated, human-curated evaluation harness specifying practical success criteria. Beyond measuring execution and task success, we also propose the alpha-value metric to quantify the economic benefit of agent performance, which integrates task success rates, token cost, and average developer salaries. Experiments across three state-of-the-art agent frameworks with multiple advanced LLMs show that leveraging code repositories for complex task solving remains challenging: even the best-performing system, OpenHands+Claude 3.7, solves only 48.15% of tasks (recent progress has pushed the frontier further, with RepoMaster+Claude 3.5 achieving a new record of 62.96%). Error analysis attributes over half of failures to seemingly mundane yet critical steps like environment setup and dependency resolution, highlighting the need for more robust workflow management and increased timeout preparedness. By releasing GitTaskBench, we aim to drive progress and attention toward repository-aware code reasoning, execution, and deployment -- moving agents closer to solving complex, end-to-end real-world tasks. The benchmark and code are open-sourced at https://github.com/QuantaAlpha/GitTaskBench.

cs.SE↗

arXiv2025

Nonreciprocal Entanglement by Dynamically Encircling a Nexus

Nonreciprocal entanglement, characterized by inherently robust operation, is a cornerstone for quantum information processing and communications. However, it remains a great challenge to achieve nonreciprocal entanglement characterized by stability and robustness against environmental fluctuations. Here, we propose a universal nonlinear mechanism to engineer magnetic-free nonreciprocity in dissipative optomechanics by utilizing bistability, a phenomenon ubiquitous across nonlinear physical systems. By dynamically encircling the nexus of bistability, a cusp converged by the bistable surfaces, we obtain nonreciprocal displacement and then utilize it to achieve robust nonreciprocal entanglement. Owing to the unique landscape of bistability, our nonreciprocal displacement and entanglements exhibit stability and robustness through closed-loop operations. Our work presents a foundational framework for leveraging nonlinearity to achieve nonreciprocal quantum information processing. It paves new avenues for exploring nonreciprocal quantum information processing and designing backaction-immune quantum metrology with nonlinearity.

quant-ph↗

arXiv2025

RepoMaster: Autonomous Exploration and Understanding of GitHub Repositories for Complex Task Solving

The ultimate goal of code agents is to solve complex tasks autonomously. Although large language models (LLMs) have made substantial progress in code generation, real-world tasks typically demand full-fledged code repositories rather than simple scripts. Building such repositories from scratch remains a major challenge. Fortunately, GitHub hosts a vast, evolving collection of open-source repositories, which developers frequently reuse as modular components for complex tasks. Yet, existing frameworks like OpenHands and SWE-Agent still struggle to effectively leverage these valuable resources. Relying solely on README files provides insufficient guidance, and deeper exploration reveals two core obstacles: overwhelming information and tangled dependencies of repositories, both constrained by the limited context windows of current LLMs. To tackle these issues, we propose RepoMaster, an autonomous agent framework designed to explore and reuse GitHub repositories for solving complex tasks. For efficient understanding, RepoMaster constructs function-call graphs, module-dependency graphs, and hierarchical code trees to identify essential components, providing only identified core elements to the LLMs rather than the entire repository. During autonomous execution, it progressively explores related components using our exploration tools and prunes information to optimize context usage. Evaluated on the adjusted MLE-bench, RepoMaster achieves a 110% relative boost in valid submissions over the strongest baseline OpenHands. On our newly released GitTaskBench, RepoMaster lifts the task-pass rate from 40.7% to 62.9% while reducing token usage by 95%. Our code and demonstration materials are publicly available at https://github.com/QuantaAlpha/RepoMaster.

cs.SE↗

arXiv2025

Learning Point Cloud Representations with Pose Continuity for Depth-Based Category-Level 6D Object Pose Estimation

Category-level object pose estimation aims to predict the 6D pose and 3D size of objects within given categories. Existing approaches for this task rely solely on 6D poses as supervisory signals without explicitly capturing the intrinsic continuity of poses, leading to inconsistencies in predictions and reduced generalization to unseen poses. To address this limitation, we propose HRC-Pose, a novel depth-only framework for category-level object pose estimation, which leverages contrastive learning to learn point cloud representations that preserve the continuity of 6D poses. HRC-Pose decouples object pose into rotation and translation components, which are separately encoded and leveraged throughout the network. Specifically, we introduce a contrastive learning strategy for multi-task, multi-category scenarios based on our 6D pose-aware hierarchical ranking scheme, which contrasts point clouds from multiple categories by considering rotational and translational differences as well as categorical information. We further design pose estimation modules that separately process the learned rotation-aware and translation-aware embeddings. Our experiments demonstrate that HRC-Pose successfully learns continuous feature spaces. Results on REAL275 and CAMERA25 benchmarks show that our method consistently outperforms existing depth-only state-of-the-art methods and runs in real-time, demonstrating its effectiveness and potential for real-world applications. Our code is at https://github.com/zhujunli1993/HRC-Pose.

cs.CV↗

arXiv2025

LIRA: Inferring Segmentation in Large Multi-modal Models with Local Interleaved Region Assistance

While large multi-modal models (LMMs) demonstrate promising capabilities in segmentation and comprehension, they still struggle with two limitations: inaccurate segmentation and hallucinated comprehension. These challenges stem primarily from constraints in weak visual comprehension and a lack of fine-grained perception. To alleviate these limitations, we propose LIRA, a framework that capitalizes on the complementary relationship between visual comprehension and segmentation via two key components: (1) Semantic-Enhanced Feature Extractor (SEFE) improves object attribute inference by fusing semantic and pixel-level features, leading to more accurate segmentation; (2) Interleaved Local Visual Coupling (ILVC) autoregressively generates local descriptions after extracting local features based on segmentation masks, offering fine-grained supervision to mitigate hallucinations. Furthermore, we find that the precision of object segmentation is positively correlated with the latent related semantics of the token. To quantify this relationship and the model's potential semantic inferring ability, we introduce the Attributes Evaluation (AttrEval) dataset. Our experiments show that LIRA achieves state-of-the-art performance in both segmentation and comprehension tasks. Code will be available at https://github.com/echo840/LIRA.

cs.CV↗

arXiv2025

Real-Time Audio-Visual Speech Enhancement Using Pre-trained Visual Representations

Speech enhancement in audio-only settings remains challenging, particularly in the presence of interfering speakers. This paper presents a simple yet effective real-time audio-visual speech enhancement (AVSE) system, RAVEN, which isolates and enhances the on-screen target speaker while suppressing interfering speakers and background noise. We investigate how visual embeddings learned from audio-visual speech recognition (AVSR) and active speaker detection (ASD) contribute to AVSE across different SNR conditions and numbers of interfering speakers. Our results show concatenating embeddings from AVSR and ASD models provides the greatest improvement in low-SNR, multi-speaker environments, while AVSR embeddings alone perform best in noise-only scenarios. In addition, we develop a real-time streaming system that operates on a computer CPU and we provide a video demonstration and code repository. To our knowledge, this is the first open-source implementation of a real-time AVSE system.

eess.AS↗

arXiv2025

Atom probe tomography of hydrated biomacromolecules: preliminary results

The folding and structure of biomacromolecules depend on the 3D distributions of their constituents, which ultimately controls their functionalities and interactions with other biomacromolecules. Atom probe tomography (APT) with its unparalleled compositional sensitivity at nanoscale spatial resolution, could provide complementary information to cryo-electron microscopy, yet routine APT analysis of biomacromolecules in their native state remains challenging. Here, a ferritin solution was used as a model system. Following plunge freezing in liquid nitrogen, cryogenic lift-out and cryo-APT analysis were performed. Elements from the ferritin core and shell are detected yet particles seem destroyed. We hence demonstrate the feasibility of preparing and analyzing bulk hydrated biological samples using APT, however, the cooling was too slow to vitrify the solution. This caused irrecoverable damage to the protein shell surrounding the ferritin particles due to ice crystal formation. We report on preliminary data from high-pressure frozen (HPF) deionized (DI) water, demonstrating a proof-ofprinciple experiments that intact biomacromolecules could be analyzed through a similar workflow in the future. We report on many trials (and errors) on the use of different materials for substrates and different substrate geometries, and provide a perspective on the challenges we faced to facilitate future studies across the community.

physics.bio-ph↗

arXiv2025

Integrating LLM-Derived Multi-Semantic Intent into Graph Model for Session-based Recommendation

Session-based recommendation (SBR) is mainly based on anonymous user interaction sequences to recommend the items that the next user is most likely to click. Currently, the most popular and high-performing SBR methods primarily leverage graph neural networks (GNNs), which model session sequences as graph-structured data to effectively capture user intent. However, most GNNs-based SBR methods primarily focus on modeling the ID sequence information of session sequences, while neglecting the rich semantic information embedded within them. This limitation significantly hampers model's ability to accurately infer users' true intention. To address above challenge, this paper proposes a novel SBR approach called Integrating LLM-Derived Multi-Semantic Intent into Graph Model for Session-based Recommendation (LLM-DMsRec). The method utilizes a pre-trained GNN model to select the top-k items as candidate item sets and designs prompts along with a large language model (LLM) to infer multi-semantic intents from these candidate items. Specifically, we propose an alignment mechanism that effectively integrates the semantic intent inferred by the LLM with the structural intent captured by GNNs. Extensive experiments conducted on the Beauty and ML-1M datasets demonstrate that the proposed method can be seamlessly integrated into GNNs framework, significantly enhancing its recommendation performance.

cs.IR↗

arXiv2025

Optimal Calibrated Signaling in Digital Auctions

In digital advertising, online platforms allocate ad impressions through real-time auctions, where advertisers typically rely on autobidding agents to optimize bids on their behalf. Unlike traditional auctions for physical goods, the value of an ad impression is uncertain and depends on the unknown click-through rate (CTR). While platforms can estimate CTRs more accurately using proprietary machine learning algorithms, these estimates/algorithms remain opaque to advertisers. This information asymmetry naturally raises the following questions: how can platforms disclose information in a way that is both credible and revenue-optimal? We address these questions through calibrated signaling, where each prior-free bidder receives a private signal that truthfully reflects the conditional expected CTR of the ad impression. Such signals are trustworthy and allow bidders to form unbiased value estimates, even without access to the platform's internal algorithms. We study the design of platform-optimal calibrated signaling in the context of second-price auction. Our first main result fully characterizes the structure of the optimal calibrated signaling, which can also be computed efficiently. We show that this signaling can extract the full surplus -- or even exceed it -- depending on a specific market condition. Our second main result is an FPTAS for computing an approximately optimal calibrated signaling that satisfies an IR condition. Our main technical contributions are: a reformulation of the platform's problem as a two-stage optimization problem that involves optimal transport subject to calibration feasibility constraints on the bidders' marginal bid distributions; and a novel correlation plan that constructs the optimal distribution over second-highest bids.

cs.GT↗

arXiv2025

Many-Body Physics from Spin-Phonon Coupling in Rydberg Atom Arrays

The rapid advancement of quantum science and technology has established Rydberg atom arrays as a premier platform for exploring quantum many-body physics with exceptional precision and controllability. Traditionally, each atom is modeled as a spin degree of freedom with its spatial motion effectively frozen. This simplification has facilitated the discovery of a rich variety of novel equilibrium and non-equilibrium phases, including $\mathbb{Z}_{\text{N}}$ symmetry-breaking orders and quantum scars. In this work, we investigate the consequences of incorporating atomic vibrations in optical tweezers, which give rise to spin-phonon coupling. For systems in thermal equilibrium, we find that this coupling leads to a new symmetry-breaking phase in the weak driving limit, as a result of induced three-spin interactions. Furthermore, we show that the violation of quantum thermalization in $\mathbb{Z}_2$-ordered states is suppressed when spin-phonon coupling is introduced. Our results are readily testable in state-of-the-art Rydberg atom array experiments.

cond-mat.quant-gas↗

arXiv2025

EECD-Net: Energy-Efficient Crack Detection with Spiking Neural Networks and Gated Attention

Crack detection on road surfaces is a critical measurement technology in the instrumentation domain, essential for ensuring infrastructure safety and transportation reliability. However, due to limited energy and low-resolution imaging, smart terminal devices struggle to maintain real-time monitoring performance. To overcome these challenges, this paper proposes a multi-stage detection approach for road crack detection, EECD-Net, to enhance accuracy and energy efficiency of instrumentation. Specifically, the sophisticated Super-Resolution Convolutional Neural Network (SRCNN) is employed to address the inherent challenges of low-quality images, which effectively enhance image resolution while preserving critical structural details. Meanwhile, a Spike Convolution Unit (SCU) with Continuous Integrate-and-Fire (CIF) neurons is proposed to convert these images into sparse pulse sequences, significantly reducing power consumption. Additionally, a Gated Attention Transformer (GAT) module is designed to strategically fuse multi-scale feature representations through adaptive attention mechanisms, effectively capturing both long-range dependencies and intricate local crack patterns, and significantly enhancing detection robustness across varying crack morphologies. The experiments on the CrackVision12K benchmark demonstrate that EECD-Net achieves a remarkable 98.6\% detection accuracy, surpassing state-of-the-art counterparts such as Hybrid-Segmentor by a significant 1.5\%. Notably, the EECD-Net maintains exceptional energy efficiency, consuming merely 5.6 mJ, which is a substantial 33\% reduction compared to baseline implementations. This work pioneers a transformative approach in instrumentation-based crack detection, offering a scalable, low-power solution for real-time, large-scale infrastructure monitoring in resource-constrained environments.

cs.CV↗