Progressive Content Refinement with Decaying Reward Joint LinUCB

arXiv:2608.06750 · cs.CL, cs.AI · Submitted 2026-08-07 · Read on arXiv

Shion Ishikawa, Pablo Loyola, Young-joo Chung, Yun Ching Liu

cs.CL, cs.AI

Submitted: 2026-08-07

Updated: 2026-08-10

License: http://creativecommons.org/publicdomain/zero/1.0/

The gist: Iterative refinement has significantly enhanced Large Language Model (LLM) performance; however, existing methods ranging from feedback-based Self-Refine to traditional bandit approaches often rely

Terminology

Abstract

Iterative refinement has significantly enhanced Large Language Model (LLM) performance; however, existing methods ranging from feedback-based Self-Refine to traditional bandit approaches often rely on static options or overlook the saturation effect. This neglect leads to over-exploitation, where the continuous use of identical prompts or arms results in diminishing rewards over time. To address this challenge, we propose a novel contextual bandit algorithm that explicitly incorporates reward decay modeling. Utilizing an Expectation-Maximization (EM) algorithm, our method simultaneously estimates both arm-specific and decay parameters. Furthermore, by embedding prompts as arms, we facilitate the joint learning of arm values, distinguishing our approach from the traditional disjoint Linear Upper Confidence Bound (LinUCB) framework. Experimental results on Sentiment Reversal and GSM8K benchmarks demonstrate that our method achieves significant performance gains over strong baselines. Finally, our ablation study confirms that the integration of reward decay modeling within the bandit framework is crucial for mitigating over-exploitation and optimizing the iterative refinement process.

Sources

Related papers