Enhancing Reinforcement Learning Fine-Tuning with an Online Refiner

• URL: http://arxiv.org/abs/2603.18088v1
• Date: Wed, 18 Mar 2026 08:37:31 GMT
• Title: Enhancing Reinforcement Learning Fine-Tuning with an Online Refiner
• Abstract summary: Constraints are essential for stabilizing reinforcement learning fine-tuning (RFT) and preventing degenerate outputs.<n>We propose textitdynamic constraints that adapt to the evolving capabilities of the fine-tuned model.<n>Experiments on dialogue and code generation show that dynamic constraints outperform both KL regularization and unconstrained baselines.
• Score: 15.619469979987429
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Constraints are essential for stabilizing reinforcement learning fine-tuning (RFT) and preventing degenerate outputs, yet they inherently conflict with the optimization objective because stronger constraints limit the ability of a fine-tuned model to discover better solutions. We propose \textit{dynamic constraints} that resolve this tension by adapting to the evolving capabilities of the fine-tuned model based on the insight that constraints should only intervene when degenerate outputs occur. We implement this by using a reference model as an \textit{online refiner} that takes the response from the fine-tuned model and generates a minimally corrected version which preserves correct content verbatim while fixing errors. A supervised fine-tuning loss then trains the fine-tuned model to produce the refined output. This mechanism yields a constraint that automatically strengthens or relaxes based on output quality. Experiments on dialogue and code generation show that dynamic constraints outperform both KL regularization and unconstrained baselines, achieving substantially higher task rewards while maintaining training stability.

Related papers

• CARL: Constraint-Aware Reinforcement Learning for Planning with LLMs [25.005954937023052]
  Large Language Models (LLMs) frequently generate plans that violate task constraints, undermining their reliability in real-world applications.<n>This deficiency arises from a lack of systematic mechanisms to incorporate constraint information during the generation process.<n>We propose Constraint-Aware Reinforcement Learning (CARL), a novel RL framework designed to strengthen LLMs' intrinsic focus on constraints.
  arXiv  Detail & Related papers  (2026-07-06T09:29:57Z)
• Constrained Flow Optimization via Sequential Fine Tuning for Molecular Design [31.59334876687123]
  We present a framework for Constrained Generative Optimization (CFO)<n>CFO balances reward and constraint satisfaction by reducing the original problem to sequential fine-tuning.<n>We provide convergence guarantees for constrained generative optimization and constrained generation via CFO.
  arXiv  Detail & Related papers  (2026-05-28T22:02:57Z)
• An Integrated Fusion Framework for Ensemble Learning Leveraging Gradient Boosting and Fuzzy Rule-Based Models [59.13182819190547]
  Fuzzy rule-based models excel in interpretability and have seen widespread application across diverse fields.<n>They face challenges such as complex design specifications and scalability issues with large datasets.<n>This paper proposes an Integrated Fusion Framework that merges the strengths of both paradigms to enhance model performance and interpretability.
  arXiv  Detail & Related papers  (2025-11-11T10:28:23Z)
• Adaptive Neighborhood-Constrained Q Learning for Offline Reinforcement Learning [52.03884701766989]
  offline reinforcement learning (RL) algorithms typically impose constraints on action selection.<n>We propose a new neighborhood constraint that restricts action selection in the Bellman target to the union of neighborhoods of dataset actions.<n>We develop a simple yet effective algorithm, Adaptive Neighborhood-constrained Q learning (ANQ), to perform Q learning with target actions satisfying this constraint.
  arXiv  Detail & Related papers  (2025-11-04T13:42:05Z)
• AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Models [62.70575022567081]
  We propose AdvChain, an alignment paradigm that teaches models dynamic self-correction through adversarial CoT tuning.<n>Our work establishes a new direction for building more robust and reliable reasoning models.
  arXiv  Detail & Related papers  (2025-09-29T04:27:23Z)
• Modular Delta Merging with Orthogonal Constraints: A Scalable Framework for Continual and Reversible Model Composition [0.0]
  Existing approaches to model merging and continual learning often suffer from task interference, catastrophic forgetting, or lack of reversibility.<n>We propose Modular Delta Merging with Orthogonal Constraints (MDM-OC), a novel framework that enables scalable, interference-free, and composition of fine-tuned models.
  arXiv  Detail & Related papers  (2025-07-28T17:08:49Z)
• Enforcing Hard Linear Constraints in Deep Learning Models with Decision Rules [8.098452803458253]
  This paper proposes a model-agnostic framework for enforcing input-dependent linear equality and inequality constraints on neural network outputs.<n>The architecture combines a task network trained for prediction accuracy with a safe network trained using decision rules from the runtime and robust optimization to ensure feasibility across the entire input space.
  arXiv  Detail & Related papers  (2025-05-20T03:09:44Z)
• Attribute Controlled Fine-tuning for Large Language Models: A Case Study on Detoxification [76.14641982122696]
  We propose a constraint learning schema for fine-tuning Large Language Models (LLMs) with attribute control. We show that our approach leads to an LLM that produces fewer inappropriate responses while achieving competitive performance on benchmarks and a toxicity detection task.
  arXiv  Detail & Related papers  (2024-10-07T23:38:58Z)
• Resilient Constrained Reinforcement Learning [87.4374430686956]
  We study a class of constrained reinforcement learning (RL) problems in which multiple constraint specifications are not identified before study. It is challenging to identify appropriate constraint specifications due to the undefined trade-off between the reward training objective and the constraint satisfaction. We propose a new constrained RL approach that searches for policy and constraint specifications together.
  arXiv  Detail & Related papers  (2023-12-28T18:28:23Z)
• Outlier-robust neural network training: variation regularization meets trimmed loss to prevent functional breakdown [2.5628953713168685]
  We tackle the challenge of outlier-robust predictive modeling using highly expressive neural networks.<n>Our approach integrates two key components: (1) a transformed trimmed loss (TTL), and (2) higher-order variation regularization (HOVR), which imposes smoothness constraints on the prediction function.
  arXiv  Detail & Related papers  (2023-08-04T12:57:13Z)
• Conditional Denoising Diffusion for Sequential Recommendation [62.127862728308045]
  Two prominent generative models, Generative Adversarial Networks (GANs) and Variational AutoEncoders (VAEs) GANs suffer from unstable optimization, while VAEs are prone to posterior collapse and over-smoothed generations. We present a conditional denoising diffusion model, which includes a sequence encoder, a cross-attentive denoising decoder, and a step-wise diffuser.
  arXiv  Detail & Related papers  (2023-04-22T15:32:59Z)

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.