DMCA
An On-Line Algorithm for Dynamic Reinforcement Learning and Planning in Reactive Environments (1990)
| Venue: | In Proc. IEEE/INNS International Joint Conference on Neural Networks |
| Citations: | 15 - 10 self |
BibTeX
@INPROCEEDINGS{Schmidhuber90anon-line,
author = {Jürgen Schmidhuber},
title = {An On-Line Algorithm for Dynamic Reinforcement Learning and Planning in Reactive Environments},
booktitle = {In Proc. IEEE/INNS International Joint Conference on Neural Networks},
year = {1990},
pages = {253--258},
publisher = {IEEE Press}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
An on-line learning algorithm for reinforcement learning with continually running recurrent networks in non-stationary reactive environments is described. Various kinds of reinforcement are considered as special types of input to an agent living in the environment. The agent's only goal is to maximize the amount of reinforcement received over time. Supervised learning techniques for recurrent networks serve to construct a differentiable model of the environmental dynamics which includes a model of future reinforcement. This model is used for learning goal directed behavior in an on-line fashion. The method extends work done by Munro, Robinson and Fallside, Werbos, Widrow, and Jordan. The possibility of using the system for planning future action sequences is investigated, and this approach is compared to approaches based on temporal difference methods. A connection to `meta-learning' (learning how to learn) is noted. Introduction Consider an agent whose movements are controlled by the...
Keyphrases
dynamic reinforcement learning reactive environment on-line algorithm recurrent network future reinforcement on-line learning algorithm on-line fashion differentiable model introduction consider non-stationary reactive environment future action sequence special type temporal difference method various kind environmental dynamic agent living
