Articles liés à Adaptive Representations for Reinforcement Learning

Adaptive Representations for Reinforcement Learning - Couverture souple

Whiteson, Shimon

 
9783642422317: Adaptive Representations for Reinforcement Learning
Couverture souple
ISBN 10 : 3642422314 ISBN 13 : 9783642422317
Editeur : Springer, 2014

Synopsis

This book presents new algorithms for reinforcement learning, a form of machine learning in which an autonomous agent seeks a control policy for a sequential decision task. Since current methods typically rely on manually designed solution representations, agents that automatically adapt their own representations have the potential to dramatically improve performance. This book introduces two novel approaches for automatically discovering high-performing representations. The first approach synthesizes temporal difference methods, the traditional approach to reinforcement learning, with evolutionary methods, which can learn representations for a broad class ofoptimization problems. This synthesis is accomplished by customizing evolutionarymethods to the on-line nature of reinforcement learning and using them to evolve representations for value function approximators.The second approach automatically learns representations based on piecewise-constant approximations of value functions. It begins with coarse representations and gradually refines them during learning, analyzing the current policy and value function to deduce the best refinements. This book also introduces a novel method for devising input representations. This method addresses the feature selection problem by extending an algorithm that evolves the topology and weights of neural networks such that it evolves their inputs too.In addition to introducing these new methods, this book presents extensive empirical results in multiple domains demonstrating that these techniques can substantially improve performance over methodswith manual representations.

Les informations fournies dans la section « Synopsis » peuvent faire référence à une autre édition de ce titre.

À propos de l'auteur

Diederik M. Roijers completed his master's in Computing Science at Utrecht University before obtaining his Ph.D. in Artificial Intelligence under the supervision of Shimon Whiteson and Frans A. Oliehoek at the University of Amsterdam in 2016. He then joined the University of Oxford as a postdoctoral research assistant. He was awarded a Postdoctoral Fellowship Grant from the FWO (Research Foundation - Flanders) and started as an FWO Postdoctoral Fellow at the Vrije Universiteit Brussel in October 2016. His research focuses on creating intelligent autonomous systems that assist humans in solving complex problems, especially those with multiple objectives. To this end, he focuses ondecision-theoretic planning and learning, which enable agents to use mathematical models to reason about the environments in which they operate. In the multi-objective problems he has been studying, the agents produce a set of possibly optimal policies that offer different trade-offs with respect to the objectives, to help users make an informed decision.

Shimon Whiteson studied English and Computer Science at Rice University before completing his doctorate in Computer Science under the supervision of Peter Stone at the University of Texas at Austin in 2007. He then spent eight years as an Assistant and then an Associate Professor at the University of Amsterdam before joining the University of Oxford as an Associate Professor in 2015. He was awarded an ERC Starting Grant from the European Research Council in 2014. His research focuses on artificial intelligence with the goal of designing, analyzing, and evaluating the algorithms that enable computational systems to acquire and execute intelligent behavior. He is particularly interested in machine learning, with which computers can learn from experience, and decision-theoretic planning, with which they can reason about their goals and deduce behavioral strategies that maximize their utility. In addition to theoretical work on these topics, he has in recent years also focused on applying them to practical problems in robotics and search engine optimization.

À propos de la quatrième de couverture

This book presents new algorithms for reinforcement learning, a form of machine learning in which an autonomous agent seeks a control policy for a sequential decision task. Since current methods typically rely on manually designed solution representations, agents that automatically adapt their own representations have the potential to dramatically improve performance. This book introduces two novel approaches for automatically discovering high-performing representations. The first approach synthesizes temporal difference methods, the traditional approach to reinforcement learning, with evolutionary methods, which can learn representations for a broad class ofoptimization problems. This synthesis is accomplished by customizing evolutionarymethods to the on-line nature of reinforcement learning and using them to evolve representations for value function approximators.The second approach automatically learns representations based on piecewise-constant approximations of value functions. It begins with coarse representations and gradually refines them during learning, analyzing the current policy and value function to deduce the best refinements. This book also introduces a novel method for devising input representations. This method addresses the feature selection problem by extending an algorithm that evolves the topology and weights of neural networks such that it evolves their inputs too.In addition to introducing these new methods, this book presents extensive empirical results in multiple domains demonstrating that these techniques can substantially improve performance over methodswith manual representations.

Les informations fournies dans la section « A propos du livre » peuvent faire référence à une autre édition de ce titre.

Éditeur
Springer
Date d'édition
2014
Langue
anglais
ISBN 10 
3642422314
ISBN 13 
9783642422317
Reliure
Broché
Nombre de pages
132
Coordonnées du fabricant
ProductSafety@springernature.com
ProductSafety@springernature.com

Europaplatz 3
Heidelberg
69115
Allemagne