arXiv subjects

Zhe Chen

Publications and source records attributed to Zhe Chen.

At least 73 records · Page 4Linked to original sources

arXiv2025

GAPS: A Clinically Grounded, Automated Benchmark for Evaluating AI Clinicians

Current benchmarks for AI clinician systems, often based on multiple-choice exams or manual rubrics, fail to capture the depth, robustness, and safety required for real-world clinical practice. To address this, we introduce the GAPS framework, a multidimensional paradigm for evaluating Grounding (cognitive depth), Adequacy (answer completeness), Perturbation (robustness), and Safety. Critically, we developed a fully automated, guideline-anchored pipeline to construct a GAPS-aligned benchmark end-to-end, overcoming the scalability and subjectivity limitations of prior work. Our pipeline assembles an evidence neighborhood, creates dual graph and tree representations, and automatically generates questions across G-levels. Rubrics are synthesized by a DeepResearch agent that mimics GRADE-consistent, PICO-driven evidence review in a ReAct loop. Scoring is performed by an ensemble of large language model (LLM) judges. Validation confirmed our automated questions are high-quality and align with clinician judgment (90% agreement, Cohen's Kappa 0.77). Evaluating state-of-the-art models on the benchmark revealed key failure modes: performance degrades sharply with increased reasoning depth (G-axis), models struggle with answer completeness (A-axis), and they are highly vulnerable to adversarial perturbations (P-axis) as well as certain safety issues (S-axis). This automated, clinically-grounded approach provides a reproducible and scalable method for rigorously evaluating AI clinician systems and guiding their development toward safer, more reliable clinical practice. The benchmark dataset GAPS-NSCLC-preview and evaluation code are publicly available at https://huggingface.co/datasets/AQ-MedAI/GAPS-NSCLC-preview and https://github.com/AQ-MedAI/MedicalAiBenchEval.

cs.CL↗

arXiv2025

RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis

Clinical diagnosis is a highly specialized discipline requiring both domain expertise and strict adherence to rigorous guidelines. While current AI-driven medical research predominantly focuses on knowledge graphs or natural text pretraining paradigms to incorporate medical knowledge, these approaches primarily rely on implicitly encoded knowledge within model parameters, neglecting task-specific knowledge required by diverse downstream tasks. To address this limitation, we propose Retrieval-Augmented Diagnosis (RAD), a novel framework that explicitly injects external knowledge into multimodal models directly on downstream tasks. Specifically, RAD operates through three key mechanisms: retrieval and refinement of disease-centered knowledge from multiple medical sources, a guideline-enhanced contrastive loss that constrains the latent distance between multi-modal features and guideline knowledge, and the dual transformer decoder that employs guidelines as queries to steer cross-modal fusion, aligning the models with clinical diagnostic workflows from guideline acquisition to feature extraction and decision-making. Moreover, recognizing the lack of quantitative evaluation of interpretability for multimodal diagnostic models, we introduce a set of criteria to assess the interpretability from both image and text perspectives. Extensive evaluations across four datasets with different anatomies demonstrate RAD's generalizability, achieving state-of-the-art performance. Furthermore, RAD enables the model to concentrate more precisely on abnormal regions and critical indicators, ensuring evidence-based, trustworthy diagnosis. Our code is available at https://github.com/tdlhl/RAD.

cs.LG↗

arXiv2025

Tyche: A Hybrid Computation Framework of Illumination Pattern for Satellite Beam Hopping

High-Throughput Satellites (HTS) use beam hopping to handle non-uniform and time-varying ground traffic demand. A significant technical challenge in beam hopping is the computation of effective illumination patterns. Traditional algorithms, like the genetic algorithm, require over 300 seconds to compute a single illumination pattern for just 37 cells, whereas modern HTS typically covers over 300 cells, rendering current methods impractical for real-world applications. Advanced approaches, such as multi-agent deep reinforcement learning, face convergence issues when the number of cells exceeds 40. In this paper, we introduce Tyche, a hybrid computation framework designed to address this challenge. Tyche incorporates a Monte Carlo Tree Search Beam Hopping (MCTS-BH) algorithm for computing illumination patterns and employs sliding window and pruning techniques to significantly reduce computation time. Specifically, MCTS-BH can compute one illumination pattern for 37 cells in just 12 seconds. To ensure real-time computation, we use a Greedy Beam Hopping (G-BH) algorithm, which provides a provisional solution while MCTS-BH completes its computation in the background. Our evaluation results show that MCTS-BH can increase throughput by up to 98.76%, demonstrating substantial improvements over existing solutions.

cs.NI↗

arXiv2025

Eunomia: A Multicontroller Domain Partitioning Framework in Hierarchical Satellite Network

With the rise of mega-satellite constellations, the integration of hierarchical non-terrestrial and terrestrial networks has become a cornerstone of 6G coverage enhancements. In these hierarchical satellite networks, controllers manage satellite switches within their assigned domains. However, the high mobility of LEO satellites and field-of-view (FOV) constraints pose fundamental challenges to efficient domain partitioning. Centralized control approaches face scalability bottlenecks, while distributed architectures with onboard controllers often disregard FOV limitations, leading to excessive signaling overhead. LEO satellites outside a controller's FOV require an average of five additional hops, resulting in a 10.6-fold increase in response time. To address these challenges, we propose Eunomia, a three-step domain-partitioning framework that leverages movement-aware FOV segmentation within a hybrid control plane combining ground stations and MEO satellites. Eunomia reduces control plane latency by constraining domains to FOV-aware regions and ensures single-hop signaling. It further balances traffic load through spectral clustering on a Control Overhead Relationship Graph and optimizes controller assignment via the Kuhn-Munkres algorithm. We implement Eunomia on the Plotinus emulation platform with realistic constellation parameters. Experimental results demonstrate that Eunomia reduces request loss by up to 58.3%, control overhead by up to 50.3\%, and algorithm execution time by 77.7% significantly outperforming current state-of-the-art solutions.

cs.NI↗

arXiv2025

Conservative Formulations of the Standard Enskog and Povzner Equations

This article introduces a conservative formulation of the Standard Enskog equation and the Povzner equation, both of which generalize the Boltzmann equation by incorporating the contribution of particle volume in collisions. The primary result expresses these collision integrals as the divergence with respect to the velocity variable v of a mass current. Moreover, the terms v C[f,f] and |v|^2 C[f,f], where C[f,f] denotes the Standard Enskog or Povzner collision integral, are represented as phase-space divergences (that is, divergences in both position and velocity) of corresponding momentum and energy currents. This work extends Villani's earlier result (Math. Modelling Numer. Anal. M2AN 33 (1999), 209--227) for the classical Boltzmann equation to the case of dense gases.

math.AP↗

arXiv2025

MedS$^3$: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision

Medical language models face critical barriers to real-world clinical reasoning applications. However, mainstream efforts, which fall short in task coverage, lack fine-grained supervision for intermediate reasoning steps, and rely on proprietary systems, are still far from a versatile, credible and efficient language model for clinical reasoning usage. To this end, we propose MedS3, a self-evolving framework that imparts robust reasoning capabilities to small, deployable models. Starting with 8,000 curated instances sampled via a curriculum strategy across five medical domains and 16 datasets, we use a small base policy model to conduct Monte Carlo Tree Search (MCTS) for constructing rule-verifiable reasoning trajectories. Self-explored reasoning trajectories ranked by node values are used to bootstrap the policy model via reinforcement fine-tuning and preference learning. Moreover, we introduce a soft dual process reward model that incorporates value dynamics: steps that degrade node value are penalized, enabling fine-grained identification of reasoning errors even when the final answer is correct. Experiments on eleven benchmarks show that MedS3 outperforms the previous state-of-the-art medical model by +6.45 accuracy points and surpasses 32B-scale general-purpose reasoning models by +8.57 points. Additional empirical analysis further demonstrates that MedS3 achieves robust and faithful reasoning behavior.

cs.CL↗

arXiv2025

Multi-Agent Deep Research: Training Multi-Agent Systems with M-GRPO

Multi-agent systems perform well on general reasoning tasks. However, the lack of training in specialized areas hinders their accuracy. Current training methods train a unified large language model (LLM) for all agents in the system. This may limit the performances due to different distributions underlying for different agents. Therefore, training multi-agent systems with distinct LLMs should be the next step to solve. However, this approach introduces optimization challenges. For example, agents operate at different frequencies, rollouts involve varying sub-agent invocations, and agents are often deployed across separate servers, disrupting end-to-end gradient flow. To address these issues, we propose M-GRPO, a hierarchical extension of Group Relative Policy Optimization designed for vertical Multi-agent systems with a main agent (planner) and multiple sub-agents (multi-turn tool executors). M-GRPO computes group-relative advantages for both main and sub-agents, maintaining hierarchical credit assignment. It also introduces a trajectory-alignment scheme that generates fixed-size batches despite variable sub-agent invocations. We deploy a decoupled training pipeline in which agents run on separate servers and exchange minimal statistics via a shared store. This enables scalable training without cross-server backpropagation. In experiments on real-world benchmarks (e.g., GAIA, XBench-DeepSearch, and WebWalkerQA), M-GRPO consistently outperforms both single-agent GRPO and multi-agent GRPO with frozen sub-agents, demonstrating improved stability and sample efficiency. These results show that aligning heterogeneous trajectories and decoupling optimization across specialized agents enhances tool-augmented reasoning tasks.

cs.AI↗

arXiv2025

Gentle Manipulation Policy Learning via Demonstrations from VLM Planned Atomic Skills

Autonomous execution of long-horizon, contact-rich manipulation tasks traditionally requires extensive real-world data and expert engineering, posing significant cost and scalability challenges. This paper proposes a novel framework integrating hierarchical semantic decomposition, reinforcement learning (RL), visual language models (VLMs), and knowledge distillation to overcome these limitations. Complex tasks are decomposed into atomic skills, with RL-trained policies for each primitive exclusively in simulation. Crucially, our RL formulation incorporates explicit force constraints to prevent object damage during delicate interactions. VLMs perform high-level task decomposition and skill planning, generating diverse expert demonstrations. These are distilled into a unified policy via Visual-Tactile Diffusion Policy for end-to-end execution. We conduct comprehensive ablation studies exploring different VLM-based task planners to identify optimal demonstration generation pipelines, and systematically compare imitation learning algorithms for skill distillation. Extensive simulation experiments and physical deployment validate that our approach achieves policy learning for long-horizon manipulation without costly human demonstrations, while the VLM-guided atomic skill framework enables scalable generalization to diverse tasks.

cs.RO↗

arXiv2025

PGD-based optimization of 3D bobsleigh track centerlines from 2D centerlines for simulation applications

The centerline of a bobsleigh track defines its geometry and is essential for simulation modeling. To reduce bBobsleigh training costs, leveraging the centerline of the bobsleigh track to construct a virtual environment that closely replicates real competitive settings presents a promising solution. However, publicly available centerline data are typically limited and it is imprecise to construct a training system solely based on 2-dimensional (2D) centerline. To address this practical issue, this paper proposes a method for generating a 3-dimensional (3D) track centerline based on 2D centerline data. Incorporating international track design regulations, the method formulates an optimization problem that considers total track length, height difference, slope constraints, and geometric continuity. A Projected Gradient Descent (PGD) algorithm is used to solve the optimization problem. The generated 3D centerlines are compared with real track data, and the results show that the method can reproduce realistic centerline trends from original or scaled 2D data. For the selected track segment, the relative errors in total length, height difference, and average slope are within 1.7%, 3.2% and 4.1%, respectively, for real 2D data and within 1.1%, 3.5% and 4.3% respectively for scaled data. All slope values remain within the allowable limits. Moreover, by adjusting the segmentation or modifying the weight of height difference in the cost function, various centerline styles applicable to different competitions can be generated. Under different segmentation and weight factors, the maximum errors reach up to 4.4%, 4.8%, and 9.8%, and 4.4%, 4.8%, and 10.0%, respectively. The proposed method provides a flexible and efficient tool for supporting bobsleigh track centerline design.

eess.SY↗

arXiv2025

Enabling Near-realtime Remote Sensing via Satellite-Ground Collaboration of Large Vision-Language Models

Large vision-language models (LVLMs) have recently demonstrated great potential in remote sensing (RS) tasks (e.g., disaster monitoring) conducted by low Earth orbit (LEO) satellites. However, their deployment in real-world LEO satellite systems remains largely unexplored, hindered by limited onboard computing resources and brief satellite-ground contacts. We propose Grace, a satellite-ground collaborative system designed for near-realtime LVLM inference in RS tasks. Accordingly, we deploy compact LVLM on satellites for realtime inference, but larger ones on ground stations (GSs) to guarantee end-to-end performance. Grace is comprised of two main phases that are asynchronous satellite-GS Retrieval-Augmented Generation (RAG), and a task dispatch algorithm. Firstly, we still the knowledge archive of GS RAG to satellite archive with tailored adaptive update algorithm during limited satellite-ground data exchange period. Secondly, propose a confidence-based test algorithm that either processes the task onboard the satellite or offloads it to the GS. Extensive experiments based on real-world satellite orbital data show that Grace reduces the average latency by 76-95% compared to state-of-the-art methods, without compromising inference accuracy.

cs.NI↗

arXiv2025

ChemVLM: Exploring the Power of Multimodal Large Language Models in Chemistry Area

Large Language Models (LLMs) have achieved remarkable success and have been applied across various scientific fields, including chemistry. However, many chemical tasks require the processing of visual information, which cannot be successfully handled by existing chemical LLMs. This brings a growing need for models capable of integrating multimodal information in the chemical domain. In this paper, we introduce \textbf{ChemVLM}, an open-source chemical multimodal large language model specifically designed for chemical applications. ChemVLM is trained on a carefully curated bilingual multimodal dataset that enhances its ability to understand both textual and visual chemical information, including molecular structures, reactions, and chemistry examination questions. We develop three datasets for comprehensive evaluation, tailored to Chemical Optical Character Recognition (OCR), Multimodal Chemical Reasoning (MMCR), and Multimodal Molecule Understanding tasks. We benchmark ChemVLM against a range of open-source and proprietary multimodal large language models on various tasks. Experimental results demonstrate that ChemVLM achieves competitive performance across all evaluated tasks. Our model can be found at https://huggingface.co/AI4Chem/ChemVLM-26B.

cs.LG↗

arXiv2025

DICE: Structured Reasoning in LLMs through SLM-Guided Chain-of-Thought Correction

When performing reasoning tasks with user-specific requirements, such as strict output formats, large language models (LLMs) often prioritize reasoning over adherence to detailed instructions. Fine-tuning LLMs on supervised datasets to address this is impractical due to high computational costs and limited parameter access. To tackle this, we propose DICE, a lightweight framework that guides small language models (SLMs) to refine LLMs' outputs through chain-of-thought (CoT) correction. DICE decouples the process by first prompting LLMs to generate natural language responses, then using trained SLMs to analyze and refine these outputs to meet structured output specifications. This framework preserves LLMs' broad knowledge and reasoning capabilities while ensuring the outputs conform to user demands. Specifically, DICE first constructs structured CoT adaptation datasets via a two-stage method and subsequently applies a dual-tuning strategy to fine-tune SLMs for generating structured outputs in an analyze-then-answer pattern. Experiments demonstrate that DICE improves the average format accuracy and content correctness of LLM outputs by 35.4\% and 29.4\%, respectively, achieving state-of-the-art (SOTA) performance over other competitive baselines.

cs.CL↗

arXiv2025

SHeaP: Self-Supervised Head Geometry Predictor Learned via 2D Gaussians

Accurate, real-time 3D reconstruction of human heads from monocular images and videos underlies numerous visual applications. As 3D ground truth data is hard to come by at scale, previous methods have sought to learn from abundant 2D videos in a self-supervised manner. Typically, this involves the use of differentiable mesh rendering, which is effective but faces limitations. To improve on this, we propose SHeaP (Self-supervised Head Geometry Predictor Learned via 2D Gaussians). Given a source image, we predict a 3DMM mesh and a set of Gaussians that are rigged to this mesh. We then reanimate this rigged head avatar to match a target frame, and backpropagate photometric losses to both the 3DMM and Gaussian prediction networks. We find that using Gaussians for rendering substantially improves the effectiveness of this self-supervised approach. Training solely on 2D data, our method surpasses existing self-supervised approaches in geometric evaluations on the NoW benchmark for neutral faces and a new benchmark for non-neutral expressions. Our method also produces highly expressive meshes, outperforming state-of-the-art in emotion classification.

cs.CV↗

arXiv2025

One-shot variable-ratio matching with fine balance

Variable-ratio matching is a flexible alternative to conventional $1$-to-$k$ matching for designing observational studies that emulate a target randomized controlled trial (RCT). To achieve fine balance -- that is, matching treated and control groups to have the same marginal distribution on selected covariates -- conventional approaches typically partition the data into strata based on estimated entire numbers and then perform a series of $1$-to-$k$ matches within each stratum, with $k$ determined by the stratum-specific entire number. This ``divide-and-conquer" strategy has notable limitations: (1) fine balance typically does not hold in the final pooled sample, and (2) more controls may be discarded than necessary. To address these limitations, we propose a one-shot variable-ratio matching algorithm. Our method produces designs with exact fine balance on selected covariates in the matched sample, mimicking a hypothetical RCT where units are first grouped into sets of different sizes and one unit within each set is assigned to treatment while others to control. Moreover, our method achieves comparable or superior balance across many covariates and retains more controls in the final matched design, compared to the ``divide-and-conquer" approach. We demonstrate the advantages of the proposed design over the conventional approach via simulations and using a dataset studying the effect of right heart catheterization on mortality among critically ill patients. The algorithm is implemented in the R package match2C.

stat.ME↗

arXiv2025

Evaluating Informative Cluster Size in Cluster Randomized Trials

In cluster randomized trials, the average treatment effect among individuals (i-ATE) can be different from the cluster average treatment effect (c-ATE) when informative cluster size is present, i.e., when treatment effects or participant outcomes depend on cluster size. In such scenarios, mixed-effects models and generalized estimating equations (GEEs) with exchangeable correlation structure are biased for both the i-ATE and c-ATE estimands, whereas GEEs with an independence correlation structure or analyses of cluster-level summaries are recommended in practice. However, when cluster size is non-informative, mixed-effects models and GEEs with exchangeable correlation structure can provide unbiased estimation and notable efficiency gains over other methods. Thus, hypothesis tests for informative cluster size would be useful to assess this key phenomenon under cluster randomization. In this work, we develop model-based, model-assisted, and randomization-based tests for informative cluster size in cluster randomized trials. We construct simulation studies to examine the operating characteristics of these tests, show they have appropriate Type I error control and meaningful power, and contrast them to existing model-based tests used in the observational study setting. The proposed tests are then applied to data from a recent cluster randomized trial, and practical recommendations for using these tests are discussed.

stat.ME↗

arXiv2025

Sequential Diffusion Language Models

Diffusion language models (DLMs) have strong theoretical efficiency but are limited by fixed-length decoding and incompatibility with key-value (KV) caches. Block diffusion mitigates these issues, yet still enforces a fixed block size and requires expensive training. We introduce Next Sequence Prediction (NSP), which unifies next-token and next-block prediction, enabling the model to adaptively determine the generation length at each step. When the length is fixed to 1, NSP reduces to standard next-token prediction. Building on NSP, we propose Sequential Diffusion Language Model (SDLM), which can retrofit pre-trained autoregressive language models (ALMs) at minimal cost. Specifically, SDLM performs diffusion inference within fixed-size mask blocks, but dynamically decodes consecutive subsequences based on model confidence, thereby preserving KV-cache compatibility and improving robustness to varying uncertainty and semantics across the sequence. Experiments show that SDLM matches or surpasses strong autoregressive baselines using only 3.5M training samples, while achieving 2.1 higher throughput than Qwen-2.5. Notably, the SDLM-32B model delivers even more pronounced efficiency gains, demonstrating the strong scalability potential of our modeling paradigm. Project page and codes: https://github.com/OpenGVLab/SDLM

cs.CL↗

arXiv2025

Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling

We introduce InternVL 2.5, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality. In this work, we delve into the relationship between model scaling and performance, systematically exploring the performance trends in vision encoders, language models, dataset sizes, and test-time configurations. Through extensive evaluations on a wide range of benchmarks, including multi-discipline reasoning, document understanding, multi-image / video understanding, real-world comprehension, multimodal hallucination detection, visual grounding, multilingual capabilities, and pure language processing, InternVL 2.5 exhibits competitive performance, rivaling leading commercial models such as GPT-4o and Claude-3.5-Sonnet. Notably, our model is the first open-source MLLMs to surpass 70% on the MMMU benchmark, achieving a 3.7-point improvement through Chain-of-Thought (CoT) reasoning and showcasing strong potential for test-time scaling. We hope this model contributes to the open-source community by setting new standards for developing and applying multimodal AI systems. HuggingFace demo see https://huggingface.co/spaces/OpenGVLab/InternVL

cs.CV↗

arXiv2025

Intra-DP: A High Performance Collaborative Inference System for Mobile Edge Computing

Deploying deep neural networks (DNNs) on resource-constrained mobile devices presents significant challenges, particularly in achieving real-time performance while simultaneously coping with limited computational resources and battery life. While Mobile Edge Computing (MEC) offers collaborative inference with GPU servers as a promising solution, existing approaches primarily rely on layer-wise model partitioning and undergo significant transmission bottlenecks caused by the sequential execution of DNN operations. To address this challenge, we present Intra-DP, a high-performance collaborative inference system optimized for DNN inference on MEC. Intra DP employs a novel parallel computing technique based on local operators (i.e., operators whose minimum unit input is not the entire input tensor, such as the convolution kernel). By decomposing their computations (operations) into several independent sub-operations and overlapping the computation and transmission of different sub-operations through parallel execution, Intra-DP mitigates transmission bottlenecks in MEC, achieving fast and energy-efficient inference. The evaluation demonstrates that Intra-DP reduces per-inference latency by up to 50% and energy consumption by up to 75% compared to state-of-the-art baselines, without sacrificing accuracy.

cs.NI↗