Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yang, Xinyi, Zeng, Liang, Dong, Heng, Yu, Chao, Wu, Xiaoran, Yang, Huazhong, Wang, Yu, Tambe, Milind, Wang, Tonghan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2502.12530
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911441541398528
author Yang, Xinyi
Zeng, Liang
Dong, Heng
Yu, Chao
Wu, Xiaoran
Yang, Huazhong
Wang, Yu
Tambe, Milind
Wang, Tonghan
author_facet Yang, Xinyi
Zeng, Liang
Dong, Heng
Yu, Chao
Wu, Xiaoran
Yang, Huazhong
Wang, Yu
Tambe, Milind
Wang, Tonghan
contents As humans increasingly share environments with diverse agents powered by RL, LLMs, and beyond, the ability to explain agent policies in natural language is vital for reliable coexistence. We introduce a general-purpose framework that trains explanation-generating LLMs via reinforcement learning from AI feedback, with distributional rewards generated by generative continuous normalizing flows (CNFs). CNFs capture the pluralistic and probabilistic nature of human judgments about explanations. Moreover, under mild assumptions, CNFs provably bound deviations from true human reward distributions when trained on noisy proxy rewards from LLMs. We design a specialized CNF architecture that selectively attends to linguistic cues in the decision context and explanations when generating rewards. Human and LLM evaluators find that our method delivers explanations that enable more accurate predictions of true agent decisions, exhibit greater logical soundness and actionability, and impose lower cognitive load than explanations trained with proxy LLM rewards or state-of-the-art RLHF and RLAIF baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2502_12530
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Translate Policy to Language: Flow Matching Generated Rewards for LLM Explanations
Yang, Xinyi
Zeng, Liang
Dong, Heng
Yu, Chao
Wu, Xiaoran
Yang, Huazhong
Wang, Yu
Tambe, Milind
Wang, Tonghan
Computation and Language
Machine Learning
As humans increasingly share environments with diverse agents powered by RL, LLMs, and beyond, the ability to explain agent policies in natural language is vital for reliable coexistence. We introduce a general-purpose framework that trains explanation-generating LLMs via reinforcement learning from AI feedback, with distributional rewards generated by generative continuous normalizing flows (CNFs). CNFs capture the pluralistic and probabilistic nature of human judgments about explanations. Moreover, under mild assumptions, CNFs provably bound deviations from true human reward distributions when trained on noisy proxy rewards from LLMs. We design a specialized CNF architecture that selectively attends to linguistic cues in the decision context and explanations when generating rewards. Human and LLM evaluators find that our method delivers explanations that enable more accurate predictions of true agent decisions, exhibit greater logical soundness and actionability, and impose lower cognitive load than explanations trained with proxy LLM rewards or state-of-the-art RLHF and RLAIF baselines.
title Translate Policy to Language: Flow Matching Generated Rewards for LLM Explanations
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2502.12530