Searcharxiv⌕ Search

arXiv subjects

Shuo Zhang

Publications and source records attributed to Shuo Zhang.

At least 91 records · Page 5Linked to original sources
arXiv2026

A multiwavelength study of the Galactic center black hole candidate MAXI J1744-294

For the first time in nearly a decade, a bright transient was detected in the central parsec (pc) of the Galaxy. MAXI J1744-294, or -- as it was known in its previous life -- Swift J174540.2-290037, was discovered in outburst by the MAXI telescope in January 2025. We present the results of a broadband, multi-wavelength study of MAXI J1744-294, including data from the NuSTAR, Chandra, XMM-Newton, Swift, and NICER X-ray telescopes, as well as complementary radio and near-infrared observations. We analyze the changing X-ray emission as the outburst evolved from the high/soft to the low/hard state. Using relativistic reflection features in the data, we estimate a spin of $a>0.92$ and viewing inclination $θ=28^{+3}_{-4}$ deg, consistent with the parameters measured for Swift J174540.2-290037. Based on the spectral and temporal characteristics of MAXI J1744-294, we reaffirm its classification as a candidate black hole (BH) low-mass X-ray binary (LMXB) -- the third candidate BH transient discovered within 20 arcsec of the Galactic supermassive black hole Sgr~A*. This work provides further evidence for a cusp of BH-LMXBs in the central pc of our Galaxy, as argued for in previous observational studies and suggested by analytical and theoretical work. Our ongoing multi-wavelength study, involving a complementary range of observatories and spanning different outburst states, can serve as a model for future time domain astrophysics research.

astro-ph.HE↗
arXiv2026

Spider-Sense: Intrinsic Risk Sensing for Efficient Agent Defense with Hierarchical Adaptive Screening

As large language models (LLMs) evolve into autonomous agents, their real-world applicability has expanded significantly, accompanied by new security challenges. Most existing agent defense mechanisms adopt a mandatory checking paradigm, in which security validation is forcibly triggered at predefined stages of the agent lifecycle. In this work, we argue that effective agent security should be intrinsic and selective rather than architecturally decoupled and mandatory. We propose Spider-Sense framework, an event-driven defense framework based on Intrinsic Risk Sensing (IRS), which allows agents to maintain latent vigilance and trigger defenses only upon risk perception. Once triggered, the Spider-Sense invokes a hierarchical defence mechanism that trades off efficiency and precision: it resolves known patterns via lightweight similarity matching while escalating ambiguous cases to deep internal reasoning, thereby eliminating reliance on external models. To facilitate rigorous evaluation, we introduce S$^2$Bench, a lifecycle-aware benchmark featuring realistic tool execution and multi-stage attacks. Extensive experiments demonstrate that Spider-Sense achieves competitive or superior defense performance, attaining the lowest Attack Success Rate (ASR) and False Positive Rate (FPR), with only a marginal latency overhead of 8.3\%.

cs.CR↗
arXiv2026

EvoFSM: Controllable Self-Evolution for Deep Research with Finite State Machines

While LLM-based agents have shown promise for deep research, most existing approaches rely on fixed workflows that struggle to adapt to real-world, open-ended queries. Recent work therefore explores self-evolution by allowing agents to rewrite their own code or prompts to improve problem-solving ability, but unconstrained optimization often triggers instability, hallucinations, and instruction drift. We propose EvoFSM, a structured self-evolving framework that achieves both adaptability and control by evolving an explicit Finite State Machine (FSM) instead of relying on free-form rewriting. EvoFSM decouples the optimization space into macroscopic Flow (state-transition logic) and microscopic Skill (state-specific behaviors), enabling targeted improvements under clear behavioral boundaries. Guided by a critic mechanism, EvoFSM refines the FSM through a small set of constrained operations, and further incorporates a self-evolving memory that distills successful trajectories as reusable priors and failure patterns as constraints for future queries. Extensive evaluations on five multi-hop QA benchmarks demonstrate the effectiveness of EvoFSM. In particular, EvoFSM reaches 58.0% accuracy on the DeepSearch benchmark. Additional results on interactive decision-making tasks further validate its generalization.

cs.AI↗
arXiv2026

Not All Negative Samples Are Equal: LLMs Learn Better from Plausible Reasoning

Learning from negative samples holds great promise for improving Large Language Model (LLM) reasoning capability, yet existing methods treat all incorrect responses as equally informative, overlooking the crucial role of sample quality. To address this, we propose Plausible Negative Samples (PNS), a method that synthesizes high-quality negative samples exhibiting expected format and structural coherence while ultimately yielding incorrect answers. PNS trains a dedicated model via reverse reinforcement learning (RL) guided by a composite reward combining format compliance, accuracy inversion, reward model assessment, and chain-of-thought evaluation, generating responses nearly indistinguishable from correct solutions. We further validate PNS as a plug-and-play data source for preference optimization across three backbone models on seven mathematical reasoning benchmarks. Results demonstrate that PNS consistently outperforms other negative sample synthesis methods, achieving an average improvement of 2.03% over RL-trained models.

cs.LG↗
arXiv2026

CRL-VLA: Continual Vision-Language-Action Learning

Lifelong learning is critical for embodied agents in open-world environments, where reinforcement learning fine-tuning has emerged as an important paradigm to enable Vision-Language-Action (VLA) models to master dexterous manipulation through environmental interaction. Thus, Continual Reinforcement Learning (CRL) is a promising pathway for deploying VLA models in lifelong robotic scenarios, yet balancing stability (retaining old skills) and plasticity (learning new ones) remains a formidable challenge for existing methods. We introduce CRL-VLA, a framework for continual post-training of VLA models with rigorous theoretical bounds. We derive a unified performance bound linking the stability-plasticity trade-off to goal-conditioned advantage magnitude, scaled by policy divergence. CRL-VLA resolves this dilemma via asymmetric regulation: constraining advantage magnitudes on prior tasks while enabling controlled growth on new tasks. This is realized through a simple but effective dual-critic architecture with novel Goal-Conditioned Value Formulation (GCVF), where a frozen critic anchors semantic consistency and a trainable estimator drives adaptation. Experiments on the LIBERO benchmark demonstrate that CRL-VLA effectively harmonizes these conflicting objectives, outperforming baselines in both anti-forgetting and forward adaptation.

cs.AI↗
arXiv2026

LFQA-E: Carefully Benchmarking Long-form QA Evaluation

Long-Form Question Answering (LFQA) involves generating comprehensive, paragraph-level responses to open-ended questions, which poses a significant challenge for evaluation due to the richness of information and flexible response format. Existing LFQA-evaluation benchmarks often lack reference answers and are limited in size and topic coverage, reducing their reliability. To address this gap, we introduce LFQA-E, a well-constructed, multilingual, and reference-based benchmark designed to rigorously evaluate automatic metrics for LFQA. LFQA-E comprises 1618 questions and 7323 pairwise comparisons across 15 topics, drawn from diverse sources such as online queries and examination questions, thereby enabling a comprehensive assessment of evaluation metrics. We examine five categories of metrics, encompassing 17 specific methods, using LFQA-E. The results demonstrate that none of the existing automatic metrics perform comparably to human judgments, highlighting their inability to capture the dense information in long-form responses. Furthermore, we present a detailed analysis of the failure cases and the generalization capacity of these metrics, offering insights to guide the future development of LFQA evaluation methods. The benchmark and code are available at https://github.com/YuchenFan48/LFQA-E.

cs.CL↗
arXiv2026

From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs

Diffusion Language Models (DLMs) enable fast generation, yet training large DLMs from scratch is costly. As a practical shortcut, adapting off-the-shelf Auto-Regressive (AR) model weights into a DLM could quickly equip the DLM with strong long-context generation capabilies. Prior "adaptation" attempts either modify logits or randomly grow attention masks to Full-Sequence diffusion, or simply transplant AR weights into a Block-Diffusion recipe, leaving two key questions unaddressed: where is the final destination of adaptation, and how to adapt better? For manifold benefits, we reframe the whole AR-to-DLM adaptation under the Block-Diffusion paradigm, transitioning from block size 1 to the final Block-Diffusion state. Concretely, the principled pathway of adaptation is designed as follows: we keep a context-causal path where causal attention is kept in the prefix, an efficient parallel adaptation procedure where an AR guidance is maintained, and gradual increment of the generation block size for a smoother transition. Built on these components, the adaptation is proved competitive on various models at different scales. With better adaptation, we propose NBDiff-7B that could inherit the long-context modeling and reasoning capabilities, and achieve state-of-the-art performance among the 7B-class DLMs. Codes: https://github.com/YuchuanTian/NBDiff.

cs.CL↗
arXiv2026

BrainFuse: a unified infrastructure integrating realistic biological modeling and core AI methodology

Neuroscience and artificial intelligence represent distinct yet complementary pathways to general intelligence. However, amid the ongoing boom in AI research and applications, the translational synergy between these two fields has grown increasingly elusive-hampered by a widening infrastructural incompatibility: modern AI frameworks lack native support for biophysical realism, while neural simulation tools are poorly suited for gradient-based optimization and neuromorphic hardware deployment. To bridge this gap, we introduce BrainFuse, a unified infrastructure that provides comprehensive support for biophysical neural simulation and gradient-based learning. By addressing algorithmic, computational, and deployment challenges, BrainFuse exhibits three core capabilities: (1) algorithmic integration of detailed neuronal dynamics into a differentiable learning framework; (2) system-level optimization that accelerates customizable ion-channel dynamics by up to 3,000x on GPUs; and (3) scalable computation with highly compatible pipelines for neuromorphic hardware deployment. We demonstrate this full-stack design through both AI and neuroscience tasks, from foundational neuron simulation and functional cylinder modeling to real-world deployment and application scenarios. For neuroscience, BrainFuse supports multiscale biological modeling, enabling the deployment of approximately 38,000 Hodgkin-Huxley neurons with 100 million synapses on a single neuromorphic chip while consuming as low as 1.98 W. For AI, BrainFuse facilitates the synergistic application of realistic biological neuron models, demonstrating enhanced robustness to input noise and improved temporal processing endowed by complex HH dynamics. BrainFuse therefore serves as a foundational engine to facilitate cross-disciplinary research and accelerate the development of next-generation bio-inspired intelligent systems.

cs.NE↗
arXiv2026

Idea2Story: An Automated Pipeline for Transforming Research Concepts into Complete Scientific Narratives

Autonomous scientific discovery with large language model (LLM)-based agents has recently made substantial progress, demonstrating the ability to automate end-to-end research workflows. However, existing systems largely rely on runtime-centric execution paradigms, repeatedly reading, summarizing, and reasoning over large volumes of scientific literature online. This on-the-spot computation strategy incurs high computational cost, suffers from context window limitations, and often leads to brittle reasoning and hallucination. We propose Idea2Story, a pre-computation-driven framework for autonomous scientific discovery that shifts literature understanding from online reasoning to offline knowledge construction. Idea2Story continuously collects peer-reviewed papers together with their review feedback, extracts core methodological units, composes reusable research patterns, and organizes them into a structured methodological knowledge graph. At runtime, underspecified user research intents are aligned to established research paradigms, enabling efficient retrieval and reuse of high-quality research patterns instead of open-ended generation and trial-and-error. By grounding research planning and execution in a pre-built knowledge graph, Idea2Story alleviates the context window bottleneck of LLMs and substantially reduces repeated runtime reasoning over literature. We conduct qualitative analyses and preliminary empirical studies demonstrating that Idea2Story can generate coherent, methodologically grounded, and novel research patterns, and can produce several high-quality research demonstrations in an end-to-end setting. These results suggest that offline knowledge construction provides a practical and scalable foundation for reliable autonomous scientific discovery.

cs.CE↗
arXiv2026

Domain-Aware Geometric Multimodal Learning for Multi-Domain Protein-Ligand Affinity Prediction

The accurate prediction of protein-ligand binding affinity is important for drug discovery yet remains challenging for multi-domain proteins, where inter-domain dynamics and flexible linkers govern molecular recognition. Current geometric deep learning methods typically treat proteins as monolithic graphs, failing to capture the distinct geometric and energetic signals at domain interfaces. To address this, we introduce DAGML (Domain-Aware Geometric Multimodal Learning), a hierarchical framework that explicitly models domain modularity. DAGML integrates a pre-trained protein language model with a novel domain-aware geometric encoder to distinguish intra- and inter-domain features, while a motif-centric ligand encoder captures pharmacophoric compatibility. We further curate a specialized multi-domain affinity benchmark, classifying complexes by binding topology (e.g., interface vs linker binders). Extensive experiments demonstrate that DAGML achieves a 21% reduction in MSE and a Pearson correlation of 0.726 compared to strong baselines. Ablation studies reveal that explicit modeling of domain interfaces is the primary driver of this improvement, particularly for ligands binding in the clefts between structural units. The code is available at https://github.com/jiankliu/DAGML.

q-bio.QM↗
arXiv2026

LiNUS: Lightweight Automatic Segmentation of Deep Brain Nuclei for Real-Time DBS Surgery

This paper proposes LiNUS, a lightweight deep learning framework for the automatic segmentation of the Subthalamic Nucleus (STN) in Deep Brain Stimulation (DBS) surgery. Addressing the challenges of small target volume and class imbalance in MRI data, LiNUS improves upon the U-Net architecture by introducing spectral normalization constraints, bilinear interpolation upsampling, and a multi-scale feature fusion mechanism. Experimental results on the Tsinghua DBS dataset (TT14) demonstrate that LiNUS achieves a Dice coefficient of 0.679 with an inference time of only 0.05 seconds per subject, significantly outperforming traditional manual and registration-based methods. Further validation on high-resolution data confirms the model's robustness, achieving a Dice score of 0.89. A dedicated Graphical User Interface (GUI) was also developed to facilitate real-time clinical application.

eess.IV↗
arXiv2026

ABC-Bench: Benchmarking Agentic Backend Coding in Real-World Development

The evolution of Large Language Models (LLMs) into autonomous agents has expanded the scope of AI coding from localized code generation to complex, repository-level, and execution-driven problem solving. However, current benchmarks predominantly evaluate code logic in static contexts, neglecting the dynamic, full-process requirements of real-world engineering, particularly in backend development which demands rigorous environment configuration and service deployment. To address this gap, we introduce ABC-Bench, a benchmark explicitly designed to evaluate agentic backend coding within a realistic, executable workflow. Using a scalable automated pipeline, we curated 224 practical tasks spanning 8 languages and 19 frameworks from open-source repositories. Distinct from previous evaluations, ABC-Bench require the agents to manage the entire development lifecycle from repository exploration to instantiating containerized services and pass the external end-to-end API tests. Our extensive evaluation reveals that even state-of-the-art models struggle to deliver reliable performance on these holistic tasks, highlighting a substantial disparity between current model capabilities and the demands of practical backend engineering. Our code is available at https://github.com/OpenMOSS/ABC-Bench.

cs.SE↗
arXiv2026

MemGovern: Enhancing Code Agents through Learning from Governed Human Experiences

While autonomous software engineering (SWE) agents are reshaping programming paradigms, they currently suffer from a "closed-world" limitation: they attempt to fix bugs from scratch or solely using local context, ignoring the immense historical human experience available on platforms like GitHub. Accessing this open-world experience is hindered by the unstructured and fragmented nature of real-world issue-tracking data. In this paper, we introduce MemGovern, a framework designed to govern and transform raw GitHub data into actionable experiential memory for agents. MemGovern employs experience governance to convert human experience into agent-friendly experience cards and introduces an agentic experience search strategy that enables logic-driven retrieval of human expertise. By producing 135K governed experience cards, MemGovern achieves a significant performance boost, improving resolution rates on the SWE-bench Verified by 4.65%. As a plug-in approach, MemGovern provides a solution for agent-friendly memory infrastructure.

cs.SE↗
arXiv2026

Sphinx: Benchmarking and Modeling for LLM-Driven Pull Request Review

Pull request (PR) review is essential for ensuring software quality, yet automating this task remains challenging due to noisy supervision, limited contextual understanding, and inadequate evaluation metrics. We present Sphinx, a unified framework for LLM-based PR review that addresses these limitations through three key components: (1) a structured data generation pipeline that produces context-rich, semantically grounded review comments by comparing pseudo-modified and merged code; (2) a checklist-based evaluation benchmark that assesses review quality based on structured coverage of actionable verification points, moving beyond surface-level metrics like BLEU; and (3) Checklist Reward Policy Optimization (CRPO), a novel training paradigm that uses rule-based, interpretable rewards to align model behavior with real-world review practices. Extensive experiments show that models trained with Sphinx achieve state-of-the-art performance on review completeness and precision, outperforming both proprietary and open-source baselines by up to 40\% in checklist coverage. Together, Sphinx enables the development of PR review models that are not only fluent but also context-aware, technically precise, and practically deployable in real-world development workflows. The data will be released after review.

cs.SE↗
arXiv2026

Resolving the Fe K$α$ Doublet of the Galactic Center Molecular Cloud G0.11-0.11 with XRISM

Fe K$α$ line emission from Galactic center molecular clouds can be produced either via fluorescence after illumination by an X-ray source or by cosmic ray ionization. Unparalleled high-resolution X-ray spectroscopy obtained by XRISM-Resolve for the galactic center molecular cloud G0.11-0.11 resolves its Fe K$α$ line complex for the first time, and points to a new method for discrimination between the X-ray reflection and cosmic ray ionization models. The Fe K$α$ line complex is resolved into Fe K$α_1$ at $E_{1} = 6.4040 \: \rm{keV}$ and Fe K$α_2$ at $E_{2}= 6.3910 \:\rm{keV}$. Both lines have non-instrumental FWHM of $\approx 3 \:\rm{eV}$, close to the predicted quantum mechanical width of the lines, suggesting scant other sources of line broadening other than instrumental and quantum effects. We measure a radial velocity of $v_{\rm{LSR}} = 50 \pm 12_{fit} \pm 14_{scale} \:\rm{km/s}$ for G0.11-0.11, achieving the same precision reached by radio observations of such clouds. The high-resolution spectrum tests for the presence of secondary Fe K$α$ lines, expected as a signature of cosmic ray proton/ion ionization. The absence of the secondary lines argues against the cosmic ray ionization model for G0.11-0.11. In the preferred X-ray reflection model, if the illuminating source is Sgr A$^{\star}$, the required luminosity for an X-ray outburst about 200 years ago is $L_8 \approx 10^{38} \:\rm{erg/s}$ in an $8\:\rm{keV}$-wide band at $8\:\rm{keV}$.

astro-ph.HE↗
arXiv2025

The very faint X-ray transient Swift J174610-290018 at the Galactic center

Very Faint X-ray Transients (VFXTs) are a class of X-ray binary systems that exhibit occasional outbursts with peak X-ray luminosities (L_X< 1e36 erg s^-1) much lower than typical X-ray transients. On 22nd February 2024, during its daily Galactic center monitoring, Swift-XRT detected a VFXT, 7 arcmin from Sgr A* dubbing it Swift J174610--290018. We aim to characterize the outburst that occurred in 2024, and a second, distinct outburst in 2025, to understand the nature and accretion flow properties of this new VFXT. Swift-XRT light curves are used to constrain the duration of the two events. We carried out X-ray spectral analysis exploiting XMM and NuSTAR data. We used Chandra and XMM observations of the last 25 years to constrain the quiescent luminosity of the source. During the 2024 outburst, which lasted about 50 days, the source reached a luminosity in the 2-10 keV band of L_X = 1.2e35 erg s^-1 (assuming it is located at the Galactic center). The 2025 outburst is shorter (about 5 days), and reached L_X = 9e34 erg s^-1. The spectral features of the source include an excess at 6.5-7 keV, which can be associated either with a single reflection line or with the ionized Fe XXV and XXVI lines. The same source was identified in both the XMM and Chandra catalogs of point sources (known as 4XMM J174610.7--290020). During previous detections, the source displayed luminosity levels ranging from L_X= 2e32 to L_X = 3e34 erg s^-1 between 2000 and 2010. Moreover, it exhibited a potential type I X-ray burst in 2004. The analysis of the outbursts and the potential type I burst strongly suggests the neutron star low mass X-ray binary (NS-LMXB) nature of the VFXT. The source can be described by an accretion disk corona (as has been recently proposed by the XRISM/Xtend analysis). This scenario explains the overall low luminosity of this transient and the peculiar iron lines in the spectrum.

astro-ph.HE↗
arXiv2025

The First X-Ray Polarimetry of an Eclipsing Low-Mass X-Ray Binary: Serendipitous IXPE Observation of AX J1745.6-2901

We present the first X-ray polarimetric measurement of the neutron star low-mass X-ray binary system AX J1745.6-2901 conducted by the Imaging X-ray Polarimetry Explorer (IXPE) satellite. This transient source, located within $ \sim $1.5' of the Galactic center, was observed serendipitously during a MAXI J1744-294 observation with a duration of 150 ks. The complex nature of the region in which AX J1745.6-2901 is located poses a challenge for studying its polarization. By performing a detailed analysis of the contamination from MAXI J1744-294 and the Galactic center diffuse emission, we find the source polarization degree PD = 14.7$\% \pm$ 4.0$\%$ and polarization angle PA = 122$^\circ \pm 8^\circ$. The phase-resolved analysis shows increase in polarization during the eclipse phase, with PD = 34.2$\% \pm$ 8.7$\%$, suggesting that the polarization-inducing mechanisms are of scattering nature, probably originating from disk winds.

astro-ph.HE↗
arXiv2025

Refined M-type Star Catalog from LAMOST DR10: Measurements of Radial Velocities, $T_\text{eff}$, log $g$, [M/H] and [$α$/M]

Precise stellar parameters for M-type stars, the Galaxy's most common stellar type, are crucial for numerous studies. In this work, we refined the LAMOST DR10 M-type star catalog through a two-stage process. First, we purified the catalog using techniques including deep learning and color-magnitude diagrams to remove 22,496 non-M spectra, correct 2,078 dwarf/giant classifications, and update 12,900 radial velocities. This resulted in a cleaner catalog containing 870,518 M-type spectra (820,493 dwarfs, 50,025 giants). Second, applying a label transfer strategy using values from APOGEE DR16 for parameter prediction with a ten-fold cross-validated CNN ensemble architecture, we predicted $T_\text{eff}$, $\log g$, [M/H], and [$α$/M] separately for M dwarfs and giants. The average internal errors for M dwarfs/giants are respectively: $T_\text{eff}$ 30/17 K, log $g$ 0.07/0.07 dex, [M/H] 0.07/0.05 dex, and [$α$/M] 0.02/0.02 dex. Comparison with APOGEE demonstrates external precisions of 34/14 K, 0.12/0.07 dex, 0.09/0.04 dex, and 0.03/0.02 dex for M dwarfs/giants, which represents precision improvements of over 20\% for M dwarfs and over 50\% for M giants compared to previous literature results. The catalog is available at https://nadc.china-vo.org/res/r101668/.

astro-ph.SR↗