Beyond Distillation: Pushing the Limits of Medical LLM Reasoning with Minimalist Rule-Based RL

• URL: http://arxiv.org/abs/2505.17952v1
• Date: Fri, 23 May 2025 14:27:37 GMT
• Title: Beyond Distillation: Pushing the Limits of Medical LLM Reasoning with Minimalist Rule-Based RL
• Authors: Che Liu, Haozhe Wang, Jiazhen Pan, Zhongwei Wan, Yong Dai, Fangzhen Lin, Wenjia Bai, Daniel Rueckert, Rossella Arcucci,
• Abstract summary: We present AlphaMed, the first medical LLM to show that reasoning capability can emerge purely through reinforcement learning (RL)<n>AlphaMed achieves state-of-the-art results on six medical QA benchmarks, outperforming models trained with conventional SFT+RL pipelines.<n>Our findings show that dataset informativeness is a key driver of reasoning performance, and that minimalist RL on informative, multiple-choice QA data is effective at inducing reasoning without CoT supervision.
• Score: 26.138061467729212
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Improving performance on complex tasks and enabling interpretable decision making in large language models (LLMs), especially for clinical applications, requires effective reasoning. Yet this remains challenging without supervised fine-tuning (SFT) on costly chain-of-thought (CoT) data distilled from closed-source models (e.g., GPT-4o). In this work, we present AlphaMed, the first medical LLM to show that reasoning capability can emerge purely through reinforcement learning (RL), using minimalist rule-based rewards on public multiple-choice QA datasets, without relying on SFT or distilled CoT data. AlphaMed achieves state-of-the-art results on six medical QA benchmarks, outperforming models trained with conventional SFT+RL pipelines. On challenging benchmarks (e.g., MedXpert), AlphaMed even surpasses larger or closed-source models such as DeepSeek-V3-671B and Claude-3.5-Sonnet. To understand the factors behind this success, we conduct a comprehensive data-centric analysis guided by three questions: (i) Can minimalist rule-based RL incentivize reasoning without distilled CoT supervision? (ii) How do dataset quantity and diversity impact reasoning? (iii) How does question difficulty shape the emergence and generalization of reasoning? Our findings show that dataset informativeness is a key driver of reasoning performance, and that minimalist RL on informative, multiple-choice QA data is effective at inducing reasoning without CoT supervision. We also observe divergent trends across benchmarks, underscoring limitations in current evaluation and the need for more challenging, reasoning-oriented medical QA benchmarks.

Related papers

• MedVLThinker: Simple Baselines for Multimodal Medical Reasoning [21.849783391186754]
  We present MedVLThinker, a suite of simple yet strong baselines for building reasoning-centric medical LMMs.<n>Our fully open recipe consists of: (1) systematic data curation for both text-only and image-text medical data, filtered according to varying levels of reasoning difficulty, and (2) two training paradigms:Supervised Fine-Tuning (SFT) on distilled reasoning traces and Reinforcement Learning with Verifiable Rewards (RLVR) based on final answer correctness.
  arXiv  Detail & Related papers  (2025-08-04T17:59:38Z)
• Towards Efficient Medical Reasoning with Minimal Fine-Tuning Data [30.407699113696076]
  Supervised Fine-Tuning (SFT) plays a pivotal role in adapting Large Language Models (LLMs) to specialized domains such as medical reasoning.<n>Existing methods attempt to alleviate this problem by selecting data based on sample difficulty, defined by knowledge and reasoning complexity.<n>We propose a data selection strategy, Difficulty-Influence Quadrant (DIQ), which prioritizes samples in the high-difficulty-high-influence quadrant.
  arXiv  Detail & Related papers  (2025-08-02T17:50:35Z)
• Truth in the Few: High-Value Data Selection for Efficient Multi-Modal Reasoning [71.3533541927459]
  We propose a novel data selection paradigm termed Activation Reasoning Potential (RAP)<n>RAP identifies cognitive samples by estimating each sample's potential to stimulate genuine multi-modal reasoning.<n>Our RAP method consistently achieves superior performance using only 9.3% of the training data, while reducing computational costs by over 43%.
  arXiv  Detail & Related papers  (2025-06-05T08:40:24Z)
• Unleashing the Reasoning Potential of Pre-trained LLMs by Critique Fine-Tuning on One Problem [53.3188041952701]
  We show that Critique Fine-Tuning (CFT) on only one problem can effectively unleash the reasoning potential of LLMs.<n>With just 5 GPU hours of training, Qwen-Math-7B-CFT show an average improvement of 15% on six math benchmarks and 16% on three logic reasoning benchmarks.<n>Results are comparable to or even surpass the results from RL with 20x less compute.
  arXiv  Detail & Related papers  (2025-06-03T18:35:52Z)
• Interleaved Reasoning for Large Language Models via Reinforcement Learning [22.403928213802036]
  Long chain-of-thought (CoT) enhances large language models' (LLM) reasoning capabilities.<n>We propose a novel training paradigm that uses reinforcement learning (RL) to guide reasoning LLMs to interleave thinking and answering for multi-hop questions.
  arXiv  Detail & Related papers  (2025-05-26T07:58:17Z)
• CoLoTa: A Dataset for Entity-based Commonsense Reasoning over Long-Tail Knowledge [19.34131843380852]
  We present a new dataset for Commonsense reasoning over Long-Tail entities (CoLoTa)<n>CoLoTa consists of 3,300 queries from question answering and claim verification tasks.<n>We propose CoLoTa as a novel benchmark for assessing both (i) LLM commonsense reasoning capabilities and their robustness to hallucinations on long-tail entities.
  arXiv  Detail & Related papers  (2025-04-20T02:47:18Z)
• CoT-RAG: Integrating Chain of Thought and Retrieval-Augmented Generation to Enhance Reasoning in Large Language Models [14.784841713647682]
  Chain-of-thought (CoT) reasoning boosts large language models' (LLMs) performance on complex tasks.<n>We propose CoT-RAG, a novel reasoning framework with three key designs.<n>We show significant accuracy gains--ranging from 4.0% to 44.3%--over state-of-the-art methods.
  arXiv  Detail & Related papers  (2025-04-18T07:55:09Z)
• SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models [39.551767637896404]
  This work revisits the dominant supervised fine-tuning (SFT) then reinforcement learning (RL) paradigm for training Large Vision-Language Models (LVLMs)<n>We show that SFT can significantly undermine subsequent RL by inducing pseudo reasoning paths'' imitated from expert models.<n>We introduce VLAA-Thinking, a new multimodal dataset designed to support reasoning in LVLMs.
  arXiv  Detail & Related papers  (2025-04-10T16:54:05Z)
• Exploring the Effect of Reinforcement Learning on Video Understanding: Insights from SEED-Bench-R1 [53.894789613838654]
  We introduce SEED-Bench-R1, a benchmark designed to evaluate post-training methods for MLLMs in video understanding.<n>It includes intricate real-world videos and complex everyday planning tasks in the format of multiple-choice questions.<n>Using Qwen2-VL-Instruct-7B as a base model, we compare RL with supervised fine-tuning (SFT)<n>Our detailed analysis reveals that RL enhances visual perception but often produces less coherent reasoning chains.
  arXiv  Detail & Related papers  (2025-03-31T17:55:23Z)
• MicroVQA: A Multimodal Reasoning Benchmark for Microscopy-Based Scientific Research [57.61445960384384]
  MicroVQA consists of 1,042 multiple-choice questions (MCQs) curated by biology experts across diverse microscopy modalities.<n> Benchmarking on state-of-the-art MLLMs reveal a peak performance of 53%.<n>Expert analysis of chain-of-thought responses shows perception errors are the most frequent, followed by knowledge errors and then overgeneralization errors.
  arXiv  Detail & Related papers  (2025-03-17T17:33:10Z)
• Cognitive-Mental-LLM: Evaluating Reasoning in Large Language Models for Mental Health Prediction via Online Text [0.0]
  This study evaluates structured reasoning techniques-Chain-of-Thought (CoT), Self-Consistency (SC-CoT), and Tree-of-Thought (ToT)-to improve classification accuracy across multiple mental health datasets sourced from Reddit.<n>We analyze reasoning-driven prompting strategies, including Zero-shot CoT and Few-shot CoT, using key performance metrics such as Balanced Accuracy, F1 score, and Sensitivity/Specificity.<n>Our findings indicate that reasoning-enhanced techniques improve classification performance over direct prediction, particularly in complex cases.
  arXiv  Detail & Related papers  (2025-03-13T06:42:37Z)
• Towards Widening The Distillation Bottleneck for Reasoning Models [39.22557129190619]
  Distillation--post-training on LRMs-generated data--is a straightforward yet effective method to enhance the reasoning abilities of smaller models.<n>We found that distilled long CoT data poses learning difficulty for small models and leads to the inheritance of biases.<n>We propose constructing tree-based CoT data from scratch via Monte Carlo Tree Search.
  arXiv  Detail & Related papers  (2025-03-03T12:17:36Z)
• MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs [55.20845457594977]
  Large language models (LLMs) have shown increasing capability in problem-solving and decision-making.<n>We present a process-based benchmark MR-Ben that demands a meta-reasoning skill.<n>Our meta-reasoning paradigm is especially suited for system-2 slow thinking.
  arXiv  Detail & Related papers  (2024-06-20T03:50:23Z)
• Understanding, Predicting and Better Resolving Q-Value Divergence in Offline-RL [86.0987896274354]
  We first identify a fundamental pattern, self-excitation, as the primary cause of Q-value estimation divergence in offline RL. We then propose a novel Self-Excite Eigenvalue Measure (SEEM) metric to measure the evolving property of Q-network at training. For the first time, our theory can reliably decide whether the training will diverge at an early stage.
  arXiv  Detail & Related papers  (2023-10-06T17:57:44Z)
• T-SciQ: Teaching Multimodal Chain-of-Thought Reasoning via Mixed Large Language Model Signals for Science Question Answering [59.63860993280275]
  Large Language Models (LLMs) have demonstrated exceptional performance in various Natural Language Processing (NLP) tasks. We propose a novel method termed T-SciQ that aims at teaching science question answering with LLM signals. Our approach achieves a new state-of-the-art performance on the ScienceQA benchmark, with an accuracy of 96.18%.
  arXiv  Detail & Related papers  (2023-05-05T11:56:30Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.