Asymptotics of Reinforcement Learning with Neural Networks
Stochastic Systems · 2022-03 · 12 (1) · 2-29
DOI: 10.1287/stsy.2021.0072 ↗Kurz gefasst
Fachartikel in der Zeitschrift Stochastic Systems, 2022-03. Verfasst von Justin Sirignano und 1 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 16 Literaturangaben.
Abstract (Originalsprache)
We prove that a single-layer neural network trained with the Q-learning algorithm converges in distribution to a random ordinary differential equation as the size of the model and the number of training steps become large. Analysis of the limit differential equation shows that it has a unique stationary solution that is the solution of the Bellman equation, thus giving the optimal control for the problem. In addition, we study the convergence of the limit differential equation to the stationary solution. As a by-product of our analysis, we obtain the limiting behavior of single-layer neural networks when trained on independent and identically distributed data with stochastic gradient descent under the widely used Xavier initialization.
Zitieren
APA
Sirignano, J., & Spiliopoulos, K. (2022). Asymptotics of Reinforcement Learning with Neural Networks. Stochastic Systems, 12(1), 2-29. https://doi.org/10.1287/stsy.2021.0072
Vancouver
Sirignano J, Spiliopoulos K. Asymptotics of Reinforcement Learning with Neural Networks. Stochastic Systems. 2022;12(1):2-29. doi:10.1287/stsy.2021.0072
BibTeX
@article{Sirignano2022,
author = {Sirignano, Justin and Spiliopoulos, Konstantinos},
title = {Asymptotics of Reinforcement Learning with Neural Networks},
journal = {Stochastic Systems},
year = {2022},
volume = {12},
number = {1},
pages = {2--29},
doi = {10.1287/stsy.2021.0072}
}
EUVIMED