arXiv subjects

Lin Yang

Publications and source records attributed to Lin Yang.

At least 55 records · Page 3Linked to original sources

arXiv2026

CoAuthorAI: A Human in the Loop System For Scientific Book Writing

Large language models (LLMs) are increasingly used in scientific writing but struggle with book-length tasks, often producing inconsistent structure and unreliable citations. We introduce CoAuthorAI, a human-in-the-loop writing system that combines retrieval-augmented generation, expert-designed hierarchical outlines, and automatic reference linking. The system allows experts to iteratively refine text at the sentence level, ensuring coherence and accuracy. In evaluations of 500 multi-domain literature review chapters, CoAuthorAI achieved a maximum soft-heading recall of 98%; in a human evaluation of 100 articles, the generated content reached a satisfaction rate of 82%. The book AI for Rock Dynamics generated with CoAuthorAI and Kexin Technology's LUFFA AI model has been published with Springer Nature. These results show that systematic human-AI collaboration can extend LLMs' capabilities from articles to full-length books, enabling faster and more reliable scientific publishing.

cs.CL↗

arXiv2026

Adaptive Manipulation Potential and Haptic Estimation for Tool-Mediated Interaction

Achieving human-level dexterity in contact-rich, tool-mediated manipulation remains a significant challenge due to visual occlusion and the underdetermined nature of haptic sensing. This paper introduces a parameterized Equilibrium Manifold (EM) as a unified representation for tool-mediated interaction, and develops a closed-loop framework that integrates haptic estimation, online planning, and adaptive stiffness control. We establish a physical-geometric duality using an adaptive manipulation potential incorporating a differentiable contact model, which induces the manifold's geometric structure and ensures that complex physical interactions are encapsulated as continuous operations on the EM. Within this framework, we reformulate haptic estimation as a manifold parameter estimation problem. Specifically, a hybrid inference strategy (haptic SLAM) is employed in which discrete object shapes are classified via particle filtering, while the continuous object pose is estimated using analytical gradients for efficient optimization. By continuously updating the parameters of the manipulation potential, the framework dynamically reshapes the induced EM to guide online trajectory replanning and implement uncertainty-aware impedance control, thereby closing the perception-action loop. The system is validated through simulation and over 260 real-world screw-loosening trials. Experimental results demonstrate robust identification and manipulation success in standard scenarios while maintaining accurate tracking. Furthermore, ablation studies confirm that haptic SLAM and uncertainty-aware stiffness modulation outperform fixed impedance baselines, effectively preventing jamming during tight tolerance interactions.

cs.RO↗

arXiv2026

Adapting to noise tails in private linear regression

While the traditional goal of statistics is to infer population parameters, modern practice increasingly demands protection of individual privacy. One way to address this need is to adapt classical statistical procedures into privacy-preserving algorithms. In this paper, we develop differentially private tail-robust methods for linear regression. The trade-off among bias, privacy, and robustness is controlled by a tunable robustification parameter in the Huber loss. We implement noisy clipped gradient descent for low-dimensional settings and noisy iterative hard thresholding for high-dimensional sparse models. Under sub-Gaussian errors, our method achieves near-optimal convergence rates while relaxing several assumptions required in earlier work. For heavy-tailed errors, we explicitly characterize how the non-asymptotic convergence rate depends on the moment index, privacy parameters, sample size, and intrinsic dimension. Our analysis shows how the moment index influences the choice of robustification parameters and, in turn, the resulting statistical error and privacy cost. By quantifying the interplay among bias, privacy, and robustness, we extend classical perspectives on privacy-preserving robust regression. The proposed methods are evaluated through simulations and two real datasets.

stat.ME↗

arXiv2026

MedMT-Bench: Can LLMs Memorize and Understand Long Multi-Turn Conversations in Medical Scenarios?

Large Language Models (LLMs) have demonstrated impressive capabilities across various specialist domains and have been integrated into high-stakes areas such as medicine. However, as existing medical-related benchmarks rarely stress-test the long-context memory, interference robustness, and safety defense required in practice. To bridge this gap, we introduce MedMT-Bench, a challenging medical multi-turn instruction following benchmark that simulates the entire diagnosis and treatment process. We construct the benchmark via scene-by-scene data synthesis refined by manual expert editing, yielding 400 test cases that are highly consistent with real-world application scenarios. Each test case has an average of 22 rounds (maximum of 52 rounds), covering 5 types of difficult instruction following issues. For evaluation, we propose an LLM-as-judge protocol with instance-level rubrics and atomic test points, validated against expert annotations with a human-LLM agreement of 91.94\%. We test 17 frontier models, all of which underperform on MedMT-Bench (overall accuracy below 60.00\%), with the best model reaching 59.75\%. MedMT-Bench can be an essential tool for driving future research towards safer and more reliable medical AI. The benchmark is available in https://openreview.net/attachment?id=aKyBCsPOHB&name=supplementary_material

cs.CL↗

arXiv2026

Towards Effective and Efficient Context-aware Nucleus Detection in Histopathology Whole Slide Images

Nucleus detection in histopathology whole slide images (WSIs) is crucial for a broad spectrum of clinical applications. The gigapixel size of WSIs necessitates the use of sliding window methodology for nucleus detection. However, mainstream methods process each sliding window independently, which overlooks broader contextual information and easily leads to inaccurate predictions. To address this limitation, recent studies additionally crop a large Filed-of-View (LFoV) patch centered on each sliding window to extract contextual features. However, such methods substantially increase whole-slide inference latency. In this work, we propose an effective and efficient context-aware nucleus detection approach. Specifically, instead of using LFoV patches, we aggregate contextual clues from off-the-shelf features of historically visited sliding windows, which greatly enhances the inference efficiency. Moreover, compared to LFoV patches used in previous works, the sliding window patches have higher magnification and provide finer-grained tissue details, thereby enhancing the classification accuracy. To develop the proposed context-aware model, we utilize annotated patches along with their surrounding unlabeled patches for training. Beyond exploiting high-level tissue context from these surrounding regions, we design a post-training strategy that leverages abundant unlabeled nucleus samples within them to enhance the model's context adaptability. Extensive experimental results on three challenging benchmarks demonstrate the superiority of our method.

eess.IV↗

arXiv2026

CompactRAG: Reducing LLM Calls and Token Overhead in Multi-Hop Question Answering

Retrieval-augmented generation (RAG) has become a key paradigm for knowledge-intensive question answering. However, existing multi-hop RAG systems remain inefficient, as they alternate between retrieval and reasoning at each step, resulting in repeated LLM calls, high token consumption, and unstable entity grounding across hops. We propose CompactRAG, a simple yet effective framework that decouples offline corpus restructuring from online reasoning. In the offline stage, an LLM reads the corpus once and converts it into an atomic QA knowledge base, which represents knowledge as minimal, fine-grained question-answer pairs. In the online stage, complex queries are decomposed and carefully rewritten to preserve entity consistency, and are resolved through dense retrieval followed by RoBERTa-based answer extraction. Notably, during inference, the LLM is invoked only twice in total - once for sub-question decomposition and once for final answer synthesis - regardless of the number of reasoning hops. Experiments on HotpotQA, 2WikiMultiHopQA, and MuSiQue demonstrate that CompactRAG achieves competitive accuracy while substantially reducing token consumption compared to iterative RAG baselines, highlighting a cost-efficient and practical approach to multi-hop reasoning over large knowledge corpora. The implementation is available at GitHub.

cs.CL↗

arXiv2026

Enhancing LLM-based Specification Generation via Program Slicing and Logical Deletion

Traditional formal specification generation methods are typically tailored to specific specification types, and therefore suffer from limited generality. In recent years, large language model (LLM)-based specification generation approaches have emerged, offering a new direction for improving the universality of automated specification synthesis. However, when dealing with complex control flow, LLMs often struggle to precisely generate complete specifications that cover substructures. Moreover, the distinctive verification pipelines adopted by existing approaches may incorrectly discard logically correct specifications, while verification tools alone cannot reliably identify correct specifications. To address these issues, we propose SLD-Spec, an LLM-based specification generation method that combines program slicing and logical deletion. Specifically, SLD-Spec augments the conventional specification generation framework with two key stages: (1) a program slicing stage that decomposes the target function into several smaller code slices, enabling LLMs to focus on more localized semantic structures and thereby improving specification relevance and completeness; and (2) a logical deletion stage that leverages LLMs to perform logical reasoning and filtering over candidate specifications so as to retain logically correct ones. Experimental results show that SLD-Spec consistently outperforms existing methods on datasets containing programs of varying complexity, verifying more programs and generating specifications that are more relevant and more complete. Further ablation studies indicate that program slicing mainly improves specification relevance and completeness, whereas logical deletion plays a key role in increasing verification success rates.

cs.SE↗

arXiv2026

Hallucination-Resistant Relation Extraction via Dependency-Aware Sentence Simplification and Two-tiered Hierarchical Refinement

Relation extraction (RE) enables the construction of structured knowledge for many downstream applications. While large language models (LLMs) have shown great promise in this task, they often struggle to reliably determine whether a relation exists, particularly in sentences with complex syntax or subtle semantics. For instance, we find that Qwen2.5-14B-Instruct incorrectly predicts a relation in 96.9% of NO-RELATION instances on SciERC, revealing a severe hallucination problem. To address these challenges, we propose DEPTH, a framework that integrates Dependency-aware sEntence simPlification and Two-tiered Hierarchical refinement into the relation extraction pipeline. Given a sentence and its candidate entity pairs, DEPTH operates in two stages: (1) the Grounding module extracts relations for each pair by leveraging their shortest dependency path, distilling the sentence into a minimal yet coherent relational context that reduces syntactic noise while preserving key semantics; (2) the Refinement module aggregates all local predictions and revises them based on a holistic understanding of the sentence, correcting omissions and inconsistencies. We further introduce a causality-driven reward model that mitigates reward hacking by disentangling spurious correlations, enabling robust fine-tuning via reinforcement learning with human feedback. Experiments on eight well-established benchmarks demonstrate that DEPTH reduces the average hallucination rate to 7.9% while achieving a 9.3% improvement in average F1 score over existing LLM-based extraction baselines.

cs.CL↗

arXiv2026

Beyond Basic Specifications? A Systematic Study of Logical Constructs in LLM-based Specification Generation

Formal specifications play a pivotal role in accurately characterizing program behaviors and ensuring software correctness. In recent years, leveraging large language models (LLMs) for the automatic generation of program specifications has emerged as a promising avenue for enhancing verification efficiency. However, existing research has been predominantly confined to generating specifications based on basic syntactic constructs, falling short of meeting the demands for high-level abstraction in complex program verification. Consequently, we propose incorporating logical constructs into existing LLM-based specification generation framework. Nevertheless, there remains a lack of systematic investigation into whether LLMs can effectively generate such complex constructs. To this end, we conduct an empirical study aimed at exploring the impact of various types of syntactic constructs on specification generation framework. Specifically, we define four syntactic configurations with varying levels of abstraction and perform extensive evaluations on mainstream program verification datasets, employing a diverse set of representative LLMs. Experimental results first confirm that LLMs are capable of generating valid logical constructs. Further analysis reveals that the synergistic use of logical constructs and basic syntactic constructs leads to improvements in both verification capability and robustness, without significantly increasing verification overhead. Additionally, we uncover the distinct advantages of two refinement paradigms. To the best of our knowledge, this is the first systematic work exploring the feasibility of utilizing LLMs for generating high-level logical constructs, providing an empirical basis and guidance for the future construction of automated program verification framework with enhanced abstraction capabilities.

cs.SE↗

arXiv2026

A Large and Precise All-Sky Photometric Standard Star Dataset Across More Than 200 Passbands

High-precision photometric standard stars play a key role in enabling accurate photometric calibration and advancing various fields of astronomy. However, due to limitations in calibration methods and the limited availability and underuse of high-precision reference data, existing photometric standard stars may suffer from insufficient numbers, systematic errors exceeding 10 milli-magnitude (mmag), limited photometric band coverage, or incomplete sky coverage, among other issues. To overcome these limitations, we have constructed the largest (over 200 million stars, 1000 times the widely recognized Landolt standards in the same magnitude range), most precise (better than 10 mmag), and most comprehensive (over 200 bands, nearly 40 times the coverage of traditional standards) all-sky standard stars. Based on standards, we have calibrated multiple survey datasets to mmag precision, and subsequently developed a complete sky distribution of stars for the Pan-STARRS system. This database, the BEst STars Database (BEST), is expected to pave the way for achieving mmag-level - or even higher - photometric precision in large-scale surveys, and to play a central role in shaping a high-precision astronomical measurement framework.

astro-ph.IM↗

arXiv2025

RecGPT-V2 Technical Report

Large language models (LLMs) have demonstrated remarkable potential in transforming recommender systems from implicit behavioral pattern matching to explicit intent reasoning. While RecGPT-V1 successfully pioneered this paradigm by integrating LLM-based reasoning into user interest mining and item tag prediction, it suffers from four fundamental limitations: (1) computational inefficiency and cognitive redundancy across multiple reasoning routes; (2) insufficient explanation diversity in fixed-template generation; (3) limited generalization under supervised learning paradigms; and (4) simplistic outcome-focused evaluation that fails to match human standards. To address these challenges, we present RecGPT-V2 with four key innovations. First, a Hierarchical Multi-Agent System restructures intent reasoning through coordinated collaboration, eliminating cognitive duplication while enabling diverse intent coverage. Combined with Hybrid Representation Inference that compresses user-behavior contexts, our framework reduces GPU consumption by 60% and improves exclusive recall from 9.39% to 10.99%. Second, a Meta-Prompting framework dynamically generates contextually adaptive prompts, improving explanation diversity by +7.3%. Third, constrained reinforcement learning mitigates multi-reward conflicts, achieving +24.1% improvement in tag prediction and +13.0% in explanation acceptance. Fourth, an Agent-as-a-Judge framework decomposes assessment into multi-step reasoning, improving human preference alignment. Online A/B tests on Taobao demonstrate significant improvements: +2.98% CTR, +3.71% IPV, +2.19% TV, and +11.46% NER. RecGPT-V2 establishes both the technical feasibility and commercial viability of deploying LLM-powered intent reasoning at scale, bridging the gap between cognitive exploration and industrial utility.

cs.IR↗

arXiv2025

Enhancing Interpretability of AR-SSVEP-Based Motor Intention Recognition via CNN-BiLSTM and SHAP Analysis on EEG Data

Patients with motor dysfunction show low subjective engagement in rehabilitation training. Traditional SSVEP-based brain-computer interface (BCI) systems rely heavily on external visual stimulus equipment, limiting their practicality in real-world settings. This study proposes an augmented reality steady-state visually evoked potential (AR-SSVEP) system to address the lack of patient initiative and the high workload on therapists. Firstly, we design four HoloLens 2-based EEG classes and collect EEG data from seven healthy subjects for analysis. Secondly, we build upon the conventional CNN-BiLSTM architecture by integrating a multi-head attention mechanism (MACNN-BiLSTM). We extract ten temporal-spectral EEG features and feed them into a CNN to learn high-level representations. Then, we use BiLSTM to model sequential dependencies and apply a multi-head attention mechanism to highlight motor-intention-related patterns. Finally, the SHAP (SHapley Additive exPlanations) method is applied to visualize EEG feature contributions to the neural network's decision-making process, enhancing the model's interpretability. These findings enhance real-time motor intention recognition and support recovery in patients with motor impairments.

cs.LG↗

arXiv2025

$A^2Flow:$ Automating Agentic Workflow Generation via Self-Adaptive Abstraction Operators

Large language models (LLMs) have shown strong potential in automating the design of agentic workflows. However, existing methods still rely heavily on manually predefined operators, limiting generalization and scalability. To address this issue, we propose $A^2Flow$, a fully automated framework for agentic workflow generation based on self-adaptive abstraction operators. $A^2Flow$ employs a three-stage operator extraction process: 1) Case-based Initial Operator Generation: leveraging expert demonstrations and LLM reasoning to generate case-specific operators; 2) Operator Clustering and Preliminary Abstraction: grouping similar operators across tasks to form preliminary abstractions; and 3) Deep Extraction for Abstract Execution Operators: applying long chain-of-thought prompting and multi-path reasoning to derive compact and generalizable execution operators. These operators serve as reusable building blocks for workflow construction without manual predefinition. Furthermore, we enhance node-level workflow search with an operator memory mechanism, which retains historical outputs to enrich context and improve decision-making. Experiments on general and embodied benchmarks show that $A^2Flow$ achieves a 2.4\% and 19.3\% average performance improvement and reduces resource usage by 37\% over state-of-the-art baselines. Homepage:https://github.com/pandawei-ele/A2FLOW

cs.AI↗

arXiv2025

A Specialized Large Language Model for Clinical Reasoning and Diagnosis in Rare Diseases

Rare diseases affect hundreds of millions worldwide, yet diagnosis often spans years. Convectional pipelines decouple noisy evidence extraction from downstream inferential diagnosis, and general/medical large language models (LLMs) face scarce real world electronic health records (EHRs), stale domain knowledge, and hallucinations. We assemble a large, domain specialized clinical corpus and a clinician validated reasoning set, and develop RareSeek R1 via staged instruction tuning, chain of thought learning, and graph grounded retrieval. Across multicenter EHR narratives and public benchmarks, RareSeek R1 attains state of the art accuracy, robust generalization, and stability under noisy or overlapping phenotypes. Augmented retrieval yields the largest gains when narratives pair with prioritized variants by resolving ambiguity and aligning candidates to mechanisms. Human studies show performance on par with experienced physicians and consistent gains in assistive use. Notably, transparent reasoning highlights decisive non phenotypic evidence (median 23.1%, such as imaging, interventions, functional tests) underpinning many correct diagnoses. This work advances a narrative first, knowledge integrated reasoning paradigm that shortens the diagnostic odyssey and enables auditable, clinically translatable decision support.

cs.CL↗

arXiv2025

Generalizing Robot Trajectories from Single-Context Human Demonstrations: A Probabilistic Approach

Generalizing robot trajectories from human demonstrations to new contexts remains a key challenge in Learning from Demonstration (LfD), particularly when only single-context demonstrations are available. We present a novel Gaussian Mixture Model (GMM)-based approach that enables systematic generalization from single-context demonstrations to a wide range of unseen start and goal configurations. Our method performs component-level reparameterization of the GMM, adapting both mean vectors and covariance matrices, followed by Gaussian Mixture Regression (GMR) to generate smooth trajectories. We evaluate the approach on a dual-arm pick-and-place task with varying box placements, comparing against several baselines. Results show that our method significantly outperforms baselines in trajectory success and fidelity, maintaining accuracy even under combined translational and rotational variations of task configurations. These results demonstrate that our method generalizes effectively while ensuring boundary convergence and preserving the intrinsic structure of demonstrated motions.

cs.RO↗

arXiv2025

A Saddle Point Remedy: Power of Variable Elimination in Non-convex Optimization

The proliferation of saddle points, rather than poor local minima, is increasingly understood to be a primary obstacle in large-scale non-convex optimization for machine learning. Variable elimination algorithms, like Variable Projection (VarPro), have long been observed to exhibit superior convergence and robustness in practice, yet a principled understanding of why they so effectively navigate these complex energy landscapes has remained elusive. In this work, we provide a rigorous geometric explanation by comparing the optimization landscapes of the original and reduced formulations. Through a rigorous analysis based on Hessian inertia and the Schur complement, we prove that variable elimination fundamentally reshapes the critical point structure of the objective function, revealing that local maxima in the reduced landscape are created from, and correspond directly to, saddle points in the original formulation. Our findings are illustrated on the canonical problem of non-convex matrix factorization, visualized directly on two-parameter neural networks, and finally validated in training deep Residual Networks, where our approach yields dramatic improvements in stability and convergence to superior minima. This work goes beyond explaining an existing method; it establishes landscape simplification via saddle point transformation as a powerful principle that can guide the design of a new generation of more robust and efficient optimization algorithms.

cs.LG↗

arXiv2025

Stellar Loci. IX. Estimation of Stellar Parameters from CSST-like Photometry

The China Space Station Telescope (CSST) will conduct a deep and wide imaging survey in the NUV-, u-, g-, r-, i-, z-, and y-bands. In this work, using theoretical data synthesized from the BOSZ spectra of Bohlin et al. (2017), along with observational data constructed from different sources, we present two methods for estimating stellar parameters from CSST-like photometry. One approach is to estimate metallicity [M/H] and surface gravity log g simultaneously by using the metallicity- and log g-dependent stellar loci. Tests with theoretical data (without photometric errors) result in precisions of 0.088 dex and 0.083 dex for [M/H] and log g, respectively. With 0.01 mag photometric errors, precision is degraded by about a factor of two, due to degeneracy in [M/H] and log g. Tests with observational data, although with larger photometric errors, result in precisions of 0.10 dex and 0.39 dex for [Fe/H] and log g, respectively, thanks to the strong correlation between stellar colors and log g in real data. The other approach is the giant-dwarf loci method to obtain classifications and metallicity estimates. With the same observational data, it achieves a better [Fe/H] precision of 0.084 dex, due to the stronger constraints imposed on log g. The method also performs well in distinguishing giants from dwarfs, particularly for red or metal-poor giants. This work demonstrates the clear potential of the CSST data, paving the way for stellar-parameter estimates for many billions of stars.

astro-ph.SR↗

arXiv2025

CPathAgent: An Agent-based Foundation Model for Interpretable High-Resolution Pathology Image Analysis Mimicking Pathologists' Diagnostic Logic

Recent advances in computational pathology have led to the emergence of numerous foundation models. These models typically rely on general-purpose encoders with multi-instance learning for whole slide image (WSI) classification or apply multimodal approaches to generate reports directly from images. However, these models cannot emulate the diagnostic approach of pathologists, who systematically examine slides at low magnification to obtain an overview before progressively zooming in on suspicious regions to formulate comprehensive diagnoses. Instead, existing models directly output final diagnoses without revealing the underlying reasoning process. To address this gap, we introduce CPathAgent, an innovative agent-based approach that mimics pathologists' diagnostic workflow by autonomously navigating across WSI based on observed visual features, thereby generating substantially more transparent and interpretable diagnostic summaries. To achieve this, we develop a multi-stage training strategy that unifies patch-level, region-level, and WSI-level capabilities within a single model, which is essential for replicating how pathologists understand and reason across diverse image scales. Additionally, we construct PathMMU-HR2, the first expert-validated benchmark for large region analysis. This represents a critical intermediate scale between patches and whole slides, reflecting a key clinical reality where pathologists typically examine several key large regions rather than entire slides at once. Extensive experiments demonstrate that CPathAgent consistently outperforms existing approaches across benchmarks at three different image scales, validating the effectiveness of our agent-based diagnostic approach and highlighting a promising direction for computational pathology.

cs.CV↗