Skill Reuse as Compression in Agentic RL

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xu, Zhikun, Feng, Yu, Dineen, Jacob, Shi, Taiwei, Zhao, Jieyu, Zhou, Ben
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911733504802816
author Xu, Zhikun
Feng, Yu
Dineen, Jacob
Shi, Taiwei
Zhao, Jieyu
Zhou, Ben
author_facet Xu, Zhikun
Feng, Yu
Dineen, Jacob
Shi, Taiwei
Zhao, Jieyu
Zhou, Ben
contents Large language model agents trained with reinforcement learning (RL) often learn brittle, task-specific shortcuts. We hypothesize that agents generalize better when their successful trajectories are structurally compressible, decomposed into a small set of reusable abstract patterns. To formalize this, we introduce ReuseRL, which grounds agentic RL in the Minimum Description Length (MDL) principle. ReuseRL extracts a shared skill dictionary from successful trajectories and augments the RL objective with a segmentation cost, explicitly penalizing idiosyncratic behaviors that encode poorly. We prove a PAC-Bayes generalization bound for this compression penalty. Across ALFWorld, TextWorld-Cooking, and Countdown-Stepwise, ReuseRL improves in- and out-of-distribution success over vanilla GRPO and strong round-length baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2605_31509
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Skill Reuse as Compression in Agentic RL
Xu, Zhikun
Feng, Yu
Dineen, Jacob
Shi, Taiwei
Zhao, Jieyu
Zhou, Ben
Machine Learning
Artificial Intelligence
Large language model agents trained with reinforcement learning (RL) often learn brittle, task-specific shortcuts. We hypothesize that agents generalize better when their successful trajectories are structurally compressible, decomposed into a small set of reusable abstract patterns. To formalize this, we introduce ReuseRL, which grounds agentic RL in the Minimum Description Length (MDL) principle. ReuseRL extracts a shared skill dictionary from successful trajectories and augments the RL objective with a segmentation cost, explicitly penalizing idiosyncratic behaviors that encode poorly. We prove a PAC-Bayes generalization bound for this compression penalty. Across ALFWorld, TextWorld-Cooking, and Countdown-Stepwise, ReuseRL improves in- and out-of-distribution success over vanilla GRPO and strong round-length baselines.
title Skill Reuse as Compression in Agentic RL
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.31509