Saved in:
Bibliographic Details
Main Author: Mercier, Jean-Marc
Format: Recurso digital
Language:
Published: Zenodo 2025
Online Access:https://doi.org/10.5281/zenodo.15163410
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • <pre>We propose a self-contained, detailed, description of a scalable standardized kernel (RKHS) approach to popular reinforcement learning algorithms, where agents interact with environments having continuous states and discrete actions spaces, dealing with possibly unstructured data. These algorithms, namely Q-Learning, Actor Critic, Q-Value Gradient, Hamilton-Jacobi-Bellman (HJB) and Heuristic Controls, are implemented with a RKHS library (\cite{codpy}) using default settings. We show that this approach to reinforcement learning is accurate, robust, sample efficient and versatile, as we benchmark our algorithms in this paper on simple games and use them as a baseline for our applications.</pre>