Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models

• URL: http://arxiv.org/abs/2603.24844v1
• Date: Wed, 25 Mar 2026 22:20:25 GMT
• Title: Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models
• Abstract summary: This paper describes a multi-answer reinforcement learning approach for training LMs to perform distributional reasoning over multiple answers during inference.<n>Across question-answering, medical diagnostic, and coding benchmarks, we observe improved diversity, coverage, and set-level calibration scores compared to single answer trained baselines.
• Score: 78.68818219506313
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Given a question, a language model (LM) implicitly encodes a distribution over possible answers. In practice, post-training procedures for LMs often collapse this distribution onto a single dominant mode. While this is generally not a problem for benchmark-style evaluations that assume one correct answer, many real-world tasks inherently involve multiple valid answers or irreducible uncertainty. Examples include medical diagnosis, ambiguous question answering, and settings with incomplete information. In these cases, we would like LMs to generate multiple plausible hypotheses, ideally with confidence estimates for each one, and without computationally intensive repeated sampling to generate non-modal answers. This paper describes a multi-answer reinforcement learning approach for training LMs to perform distributional reasoning over multiple answers during inference. We modify the RL objective to enable models to explicitly generate multiple candidate answers in a single forward pass, internalizing aspects of inference-time search into the model's generative process. Across question-answering, medical diagnostic, and coding benchmarks, we observe improved diversity, coverage, and set-level calibration scores compared to single answer trained baselines. Models trained with our approach require fewer tokens to generate multiple answers than competing approaches. On coding tasks, they are also substantially more accurate. These results position multi-answer RL as a principled and compute-efficient alternative to inference-time scaling procedures such as best-of-k. Code and more information can be found at https://multi-answer-rl.github.io/.

Related papers

• Revisiting Greedy Decoding for Visual Question Answering: A Calibration Perspective [16.787456677570127]
  Greedy sampling strategies are widely adopted in large language models to balance output coherence and diversity.<n>We show that greedy decoding can be an efficient yet strong default for Visual Questioning (VQA)<n>We propose Reasoning Models, which outperforms both sampling and standard decoding in multimodal reasoning scenarios.
  arXiv  Detail & Related papers  (2026-04-25T21:01:05Z)
• Test-Time Scaling with Diffusion Language Models via Reward-Guided Stitching [66.39914384073145]
  We propose a self-consistency framework that turns cheap diffusion-sampled reasoning into a reusable pool of step-level candidates.<n>We find that step-level recombination is most beneficial on harder problems.<n>Our training-free framework improves average accuracy by up to 2 across six math and coding tasks.
  arXiv  Detail & Related papers  (2026-02-26T11:08:39Z)
• No Compute Left Behind: Rethinking Reasoning and Sampling with Masked Diffusion Models [42.158430350703505]
  Masked diffusion language models are trained to in-fill positions in randomly masked sequences.<n>We propose reasoning-as-infilling and multi-token entropy decoding.<n>Our work demonstrates that the training and compute used by MDLMs unlock many new inference and post-training methods.
  arXiv  Detail & Related papers  (2025-10-22T19:41:27Z)
• The Majority is not always right: RL training for solution aggregation [53.1050856072799]
  We train an aggregator model to review, reconcile, and synthesize a final, correct answer.<n>A key ingredient is careful balancing of easy and hard training examples.<n>We find our method, AggLM, outperforms both strong rule-based and reward-model baselines.
  arXiv  Detail & Related papers  (2025-09-08T16:39:38Z)
• Learning to Refine: Self-Refinement of Parallel Reasoning in LLMs [102.48588475875749]
  We introduce Generative Self-Refinement (GSR), a novel parallel test-time scaling framework.<n>GSR generates a set of candidate responses in parallel and then performs self-refinement to synthesize a new superior solution.<n>We show that our method achieves state-of-the-art performance across five mathematical benchmarks.
  arXiv  Detail & Related papers  (2025-08-27T06:51:48Z)
• Learning to Reason Across Parallel Samples for LLM Reasoning [48.41933431325965]
  Scaling test-time compute brings substantial performance gains for large language models (LLMs)<n>In this paper, we propose a new way to leverage such multiple sample set.<n> Experiments on five reasoning datasets demonstrate both the efficacy and efficiency of SSA.
  arXiv  Detail & Related papers  (2025-06-10T17:42:35Z)
• DCR: Divide-and-Conquer Reasoning for Multi-choice Question Answering with LLMs [9.561022942046279]
  We propose Divide and Conquer Reasoning (DCR) to enhance the reasoning capability of large language models (LLMs) We first categorize questions into two subsets based on confidence score ($mathcalCS$), which is estimated by statistical frequency of generated answers. In particular, we first categorize questions into two subsets based on confidence score ($mathcalCS$), which is estimated by statistical frequency of generated answers.
  arXiv  Detail & Related papers  (2024-01-10T14:38:46Z)
• Generative Judge for Evaluating Alignment [84.09815387884753]
  We propose a generative judge with 13B parameters, Auto-J, designed to address these challenges. Our model is trained on user queries and LLM-generated responses under massive real-world scenarios. Experimentally, Auto-J outperforms a series of strong competitors, including both open-source and closed-source models.
  arXiv  Detail & Related papers  (2023-10-09T07:27:15Z)
• GRACE: Discriminator-Guided Chain-of-Thought Reasoning [75.35436025709049]
  We propose Guiding chain-of-thought ReAsoning with a CorrectnEss Discriminator (GRACE) to steer the decoding process towards producing correct reasoning steps. GRACE employs a discriminator trained with a contrastive loss over correct and incorrect steps, which is used during decoding to score next-step candidates.
  arXiv  Detail & Related papers  (2023-05-24T09:16:51Z)
• ManyModalQA: Modality Disambiguation and QA over Diverse Inputs [73.93607719921945]
  We present a new multimodal question answering challenge, ManyModalQA, in which an agent must answer a question by considering three distinct modalities. We collect our data by scraping Wikipedia and then utilize crowdsourcing to collect question-answer pairs.
  arXiv  Detail & Related papers  (2020-01-22T14:39:28Z)

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.