The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping

TL;DR

This paper introduces MEDS, a memory-enhanced reward shaping method that uses historical behavioral data to penalize recurring errors, thereby improving diversity and performance in reinforcement learning for language models.

Contribution

MEDS is a novel framework that incorporates past behavioral signals into reward design to reduce repeated mistakes and enhance exploration in language model training.

Findings

MEDS improves performance by up to 4.13 pass@1 points.

MEDS increases behavioral diversity during sampling.

Consistent gains across five datasets and three models.

Abstract

Despite the success of reinforcement learning for large language models, a common failure mode is reduced sampling diversity, where the policy repeatedly generates similar erroneous behaviors. Classical entropy regularization encourages randomness under the current policy, but does not explicitly discourage recurrent failure patterns across rollouts. We propose MEDS, a Memory-Enhanced Dynamic reward Shaping framework that incorporates historical behavioral signals into reward design. By storing and leveraging intermediate model representations, we capture features of past rollouts and use density-based clustering to identify frequently recurring error patterns. Rollouts assigned to more prevalent error clusters are penalized more heavily, encouraging broader exploration while reducing repeated mistakes. Across five datasets and three base models, MEDS consistently improves average performance over existing baselines, achieving gains of up to 4.13 pass@1 points and 4.37 pass@128 points. Additional analyses using both LLM-based annotations and quantitative diversity metrics show that MEDS increases behavioral diversity during sampling.