Understanding Policy Gradient Algorithms: A Sensitivity-Based Approach
Shuang Wu, Ling Shi, Jun Wang, Guangjian Tian
Abstract
The REINFORCE algorithm \cite{williams1992simple} is popular in policy gradient (PG) for solving reinforcement learning (RL) problems. Meanwhile, the theoretical form of PG is from \cite{sutton1999policy}. Although both formulae prescribe PG, their precise connections are not yet illustrated. Recently, \citeauthor{nota2020policy} (\citeyear{nota2020policy}) have found that the ambiguity causes implementation errors. Motivated by the ambiguity and implementation incorrectness, we study PG from a perturbation perspective. In particular, we derive PG in a unified framework, precisely clarify the relation between PG implementation and theory, and echos back the findings by \citeauthor{nota2020policy}. Diving into factors contributing to empirical successes of the existing erroneous implementations, we find that small approximation error and the experience replay mechanism play critical roles.
BibTeX
@InProceedings{pmlr-v162-wu22i,
title = {Understanding Policy Gradient Algorithms: A Sensitivity-Based Approach},
author = {Wu, Shuang and Shi, Ling and Wang, Jun and Tian, Guangjian},
booktitle = {Proceedings of the 39th International Conference on Machine Learning},
pages = {24131--24149},
year = {2022},
editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},
volume = {162},
series = {Proceedings of Machine Learning Research},
month = {17--23 Jul},
publisher = {PMLR},
pdf = {https://proceedings.mlr.press/v162/wu22i/wu22i.pdf},
url = {https://proceedings.mlr.press/v162/wu22i.html},
abstract = {The REINFORCE algorithm \cite{williams1992simple} is popular in policy gradient (PG) for solving reinforcement learning (RL) problems. Meanwhile, the theoretical form of PG is from \cite{sutton1999policy}. Although both formulae prescribe PG, their precise connections are not yet illustrated. Recently, \citeauthor{nota2020policy} (\citeyear{nota2020policy}) have found that the ambiguity causes implementation errors. Motivated by the ambiguity and implementation incorrectness, we study PG from a perturbation perspective. In particular, we derive PG in a unified framework, precisely clarify the relation between PG implementation and theory, and echos back the findings by \citeauthor{nota2020policy}. Diving into factors contributing to empirical successes of the existing erroneous implementations, we find that small approximation error and the experience replay mechanism play critical roles.}
}