Searcharxiv⌕ Search

arXiv subjects

Ruoxuan Shi

Publications and source records attributed to Ruoxuan Shi.

3 recordsLinked to original sources

Safety-Preserving Medical LLM Fine-Tuning via Bilevel Data Selection

As general large language models (LLMs) are increasingly adapted for patient-facing medical applications, improving medical capability while controlling patient-facing risks remains a critical challenge. Large-scale medical instruction data are essential for acquiring domain knowledge through supervised fine-tuning (SFT), but unsafe responses can undermine existing safety alignment. To address this problem, we propose Scalable Bilevel Data Selection (S-BDS) for safety-preserving medical SFT. By controlling selection-weight concentration, S-BDS enables reliable sample ranking over large candidate pools and identifies training examples that improve medical capability while controlling patient-facing risks. We combine clinician-validated safe data and selected data with unscreened medical instruction data for SFT, and jointly evaluate medical capability and patient-facing safety on HealthBench and PatientSafetyBench, respectively. Experiments across Llama and Qwen models show that S-BDS achieves the highest safe rates among all compared methods, outperforming the strongest baselines by 4.64 and 3.00 percentage points, respectively, while maintaining competitive medical capability. Further experiments show that varying the safety-oriented weight during SFT controls the trade-off between medical capability and patient-facing safety.

cs.CL↗

Aligning Information Theory and Formal Verification: A Lean Formalization of Rate-Distortion Theory

Large language models have recently demonstrated an ability to generate substantial mathematical proofs and formalizations, including machine-checked arguments for major open problems and classical theorems. These developments highlight a growing workflow in which AI systems search for proofs while proof assistants certify their correctness. Within information theory, however, formalized libraries remain limited, and existing formalized theorems focus almost exclusively on Shannon's foundational lossless theorems. This paper presents the first Lean formalization of core components of rate-distortion theory, a central framework for lossy compression and modern machine-learning analysis. We identify the alignment challenges that arise when translating information-theoretic definitions, such as distortion measures, typicality and achievability, into a proof-assistant setting, and we develop a reusable library of definitions and lemmas faithful to textbook conventions. Our formalization provides a foundation for future AI-assisted proofs in information theory and enables subsequent mechanized developments to build on a coherent, verified infrastructure rather than generating long proofs from scratch.

cs.IT↗

ABO-Med: Accelerated Bilevel Optimization for Few-Shot Medical Image Classification

In recent years, bilevel optimization has been widely used in a variety of machine learning tasks. However, prior bilevel optimization algorithms generally require the computation of second-order information, which limits their practical scalability. Only recently has a first-order paradigm for bilevel optimization been established, attaining near-optimal theoretical guarantees for solving bilevel optimization problems. In this paper, we propose ABO-Med, a scalable instantiation of this paradigm for few-shot learning, by incorporating it into the model-agnostic meta-learning (MAML) framework and tailoring it to medical image classification. We also introduce Medical Adaptive RandomAugment (MedRAug), a modality-aware augmentation strategy designed for medical images. Theoretically, ABO-Med establishes the optimality of MAML-type meta-learning approaches. Empirically, ABO-Med outperforms prior baselines on several public medical datasets, with gains of 1.99% to 18.76%, while MedRAug further improves the average accuracy by 2.20% to 6.34%. Additional cross-domain experiments, augmentation ablation studies, backbone ablation studies, and training efficiency analysis further validate the effectiveness and efficiency of the proposed method.

cs.CV↗