通过重新定义谓词进行反向解释

Aug, 2024

Backward explanations via redefinition of predicates

Léo Saulières, Martin C. Cooper, Florence Dupin de Saint Cyr

TL;DR本研究针对历史解释基于谓词（HXP）面临的长历史操作重要性评分计算的复杂性问题，提出了一种新的方法——反向HXP（Backward-HXP），以无须近似评分的方式提供解释。实验结果表明，反向HXP能够有效地总结长历史，提升了解释质量。

Abstract

History Explanation based on Predicates (HXP), studies the behavior of a Reinforcement Learning (RL) agent in a sequence of agent's interactions with the environment (a history), through the prism of an arbitrary