arXiv subjects

Zhiyun Zhang

Publications and source records attributed to Zhiyun Zhang.

5 recordsLinked to original sources

arXiv2026

FaithMed: Training LLMs For Faithful Evidence-Based Medical Reasoning

Faithful reasoning is essential in medicine, where clinical decisions require transparent justification grounded in reliable evidence. Current medical LLMs either lack active access to evidence or use retrieved evidence without supervising how it should be appraised and applied during reasoning. To address this, we formalize evidence-based medicine principles as process-level criteria and introduce FaithMed, a framework that combines clinician-designed, automatically refined rubrics with reinforcement learning using step-level process reward assignment and advantage grouping. Across seven medical benchmarks, FaithMed improves over agentic-search baselines (+9% on average) and outcome-only RL (+5.8%), while raising average evidence-based medicine rubric scores over agentic-search Qwen3 baselines (+15.5%). This work demonstrates that explicit step-level supervision can improve both task success and the faithfulness of the reasoning process. Code is available at https://github.com/cxcscmu/FaithMed.

cs.CL↗

arXiv2026

Dual-Correction Physics-Informed Neural Networks for Hemodynamic Reconstruction from Sparse Data

Quantifying hemodynamics in the curved segments of the intracranial internal carotid artery is a core challenge in diagnosing vascular stenosis. Conventional full-field imaging, such as 4D Flow MRI, is costly and difficult to widely promote. Meanwhile, reconstructing full-field fluid information from easily accessible and non-invasive sparse measurement data (such as transcranial Doppler ultrasound/computed tomography angiography) is essentially a highly challenging ill-posed inverse problem. To overcome the severe optimization difficulties and generalization failures of conventional physics-informed neural networks (PINNs) in highly tortuous geometries, we propose a dual-correction physics-informed neural network (DCP-INN) framework taking into account a causal decoupling strategy. The proposed DCP-INN model utilizes a diamond-shaped main network to capture low-frequency trends in physical evolution, and employs a parallel wide-deep correction network to compensate for high-frequency residuals resulting from complex geometric shapes. Furthermore, the framework introduces a high-order physical loss function based on Taylor expansion to enhance local continuity under extremely sparse data constraints. To validate the proposed method, we performed computational evaluations on realistic vascular geometries with significant tortuosity. The results demonstrate that the method effectively mitigates optimization challenges and significantly reduces flow field reconstruction error. This study not only achieves physically credible and robust flow field reconstruction in complex morphologies but also provides a highly promising algorithmic foundation for building low-cost, high-resolution personalized cardiovascular digital twins in future.

physics.med-ph↗

arXiv2026

Astrometric properties of reference frame sources as a function of redshift

Previous studies based on the latest realisation of the International Celestial Reference Frame (ICRF3) have suggested a correlation between astrometric properties (such as the radio-optical offset) and redshift for active galactic nuclei (AGNs). We extend these investigations by using a large, all-sky sample of approximately 22,000 compact radio sources from the Radio Fundamental Catalogue (RFC) to examine this relationship in a systematic and statistically robust manner. We compiled redshifts for about 10,000 RFC sources over the range 0 < z < 5 by combining data from the Dark Energy Spectroscopic Instrument Data Release 1 and the Sloan Digital Sky Survey Data Release 17/19 with additional datasets from the NASA/IPAC Extragalactic Database. Cross-matching with Gaia Data Release 3 yielded a sample of 4,068 RFC objects with reliable spectroscopic redshifts and classifications, including galaxies and quasi-stellar objects (QSOs). We analysed the redshift dependence of their radio astrometric properties from very long baseline interferometry (VLBI) and their optical astrometric properties from Gaia. We find that the VLBI astrometric properties show no significant dependence on redshift within the achieved level of precision. In contrast, several optical astrometric quantities exhibit clear redshift-dependent behaviour. The median absolute radio-optical offsets decrease markedly over 0 < z < 0.5, where galaxies dominate the sample, decline more gradually over 0.5 < z < 1.3, and exhibit a mild increase at z > 1.3, where QSOs dominate. Similar behaviour is observed for several Gaia astrometric quantities, including astrometric uncertainties, proper motions, and G magnitudes. These behaviours can be largely explained by the dependence of Gaia astrometric performance on G magnitude and by the evolution of the G magnitude with redshift.

astro-ph.IM↗

arXiv2026

GENERator: A Long-Context Generative Genomic Foundation Model

The rapid advancement of DNA sequencing has produced vast genomic datasets, yet interpreting and engineering genomic function remain fundamental challenges. Recent large language models have opened new avenues for genomic analysis, but existing approaches are often limited by restricted training scope, constrained generative capability, or prohibitive computational cost. We introduce GENErator, a generative genomic foundation model for long-context DNA modeling, with a context length of 98k nucleotides, pre-trained on 386 billion nucleotides of eukaryotic DNA. Without task-specific fine-tuning, GENERator exhibits strong intrinsic capabilities: unsupervised embedding analyses reveal phylogenetically coherent structure, and sequence recovery benchmarks demonstrate generative accuracy comparable to or exceeding state-of-the-art models with substantially improved computational efficiency. In a zero-shot setting, GENERator achieves competitive variant effect prediction performance relative to alignment-based methods, while remaining fully alignment-free and broadly applicable across species. With task-specific fine-tuning, the model attains leading performance on established genomic benchmarks. We further demonstrate practical generative applications. GENERator can generate protein-coding DNA sequences that translate into structurally plausible proteins and, through a prompt-guided design framework, design cis-regulatory elements with targeted activity profiles, including synthetic super-enhancers validated by high-throughput UMI-STARR-seq assays. Together, these results establish GENERator as an efficient and biologically grounded framework for genomic interpretation and programmable sequence design. Code and supplementary resources are available at https://github.com/GenerTeam/GENERator.

cs.CL↗

arXiv2025

The origin of double-peaked narrow emission-line galaxies in MaNGA Survey

We select 36 double-peaked narrow emission-line galaxies (DPGs) from 10,010 unique galaxies in MaNGA survey. These DPGs show double-peaked Balmer lines and forbidden lines in the spectra. We use a double Gaussian model to separate the double-peaked profiles of each emission line into blue and red components ($λ_\text{blue}$ < $λ_\text{red}$), and analyze the spatially resolved kinematics and ionization mechanisms of each component. We find that in 35 out of 36 DPGs, the flux ratio between the blue and red components varies systematically along the major axes, while it keeps roughly a constant along the minor axes. The blue and red components of these DPGs exhibit similar distributions in both the value of line-of-sight velocity and the velocity dispersion. Additionally, 83.3% DPGs have both blue and red components located in the same ionization region in the [SII]-BPT diagram. Combining all these observational results, we suggest that the double-peaked emission line profiles in these 35 DPGs primarily originate from rotating discs. The remaining one galaxy shows clear outflow features. 8 out of 35 DPGs show symmetric line profiles that indicate undisturbed rotating discs, and the other 27 DPGs exhibit asymmetric profiles, suggesting dynamic disturbances in the rotating discs. Furthermore, we find that 58.3% DPGs experienced external processes, characterized by tidal features, companion galaxies, as well as gas-star misalignments. This fraction is about twice as much as that of the control sample, suggesting the origin of double-peaked emission line profiles is associated with external processes.

astro-ph.GA↗