Adaptive Representations for Reinforcement Learning - download pdf or read online

By Shimon Whiteson

ISBN-10: 3642139310

ISBN-13: 9783642139314

This ebook provides new algorithms for reinforcement studying, a kind of laptop studying within which an self sufficient agent seeks a regulate coverage for a sequential selection job. due to the fact present tools generally depend upon manually designed answer representations, brokers that instantly adapt their very own representations have the capability to dramatically enhance functionality. This publication introduces novel ways for immediately researching high-performing representations. the 1st method synthesizes temporal distinction equipment, the conventional method of reinforcement studying, with evolutionary tools, that can study representations for a wide classification of optimization difficulties. This synthesis is complete via customizing evolutionary the way to the online nature of reinforcement studying and utilizing them to adapt representations for worth functionality approximators. the second one process instantly learns representations in response to piecewise-constant approximations of price capabilities. It starts with coarse representations and steadily refines them in the course of studying, interpreting the present coverage and cost functionality to infer the easiest refinements. This e-book additionally introduces a singular strategy for devising enter representations. this technique addresses the function choice challenge through extending an set of rules that evolves the topology and weights of neural networks such that it evolves their inputs too. as well as introducing those new equipment, this e-book provides broad empirical leads to a number of domain names demonstrating that those strategies can considerably enhance functionality over equipment with guide representations.

Show description

Read or Download Adaptive Representations for Reinforcement Learning PDF

Best nonfiction_6 books

New PDF release: Recognizing the Stranger: Recognition Scenes in the Gospel

"Recognizing the Stranger" is the 1st monographic research of popularity scenes and motifs within the Gospel of John. the popularity type-scene ("anagn? risis") used to be a typical characteristic in historic drama and narrative, hugely valued through Aristotle as a touching second of fact, e. g. , in Oedipus' tragic self-discovery and Odysseus' chuffed homecoming.

Extra resources for Adaptive Representations for Reinforcement Learning

Sample text

We tested both Darwinian and Lamarckian NEAT+Q in this manner. Both perform well, though which is preferable appears to be domain dependent. For simplicity, in this section and those that follow, we present results only for Darwinian NEAT+Q. 4 we present a comparison of the two approaches. To test Q-learning without NEAT, we tried 24 different configurations in each domain. These configurations correspond to every possible combination of the following parameter settings. The networks had feed-forward topologies with 0, 4, or 8 hidden nodes.

By focusing exploration on the most promising individuals, softmax and interval estimation offer the best of both worlds: they excel at the on-line metrics without sacrificing the quality of the best policies discovered. 30 3 On-Line Evolutionary Computation Overall, these results verify the efficacy of these methods of on-line evolution. It is less clear, however, which strategy is most useful. Softmax clearly outperforms ε greedy but may be more difficult to use in practice because the τ parameter is harder to tune, as evidenced by the need to assign it different values in the two domains.

6 compares the performance of Darwinian and Lamarckian NEAT+Q in both the mountain car and server job scheduling domains. In both cases, we use off-line NEAT+Q, as the on-line versions tend to mute the differences between the two implementations. Though both implementations perform well in both domains, Lamarckian NEAT+Q does better in mountain car but worse in server job scheduling. Hence, the relative performance of these two approaches seems to depend critically on the dynamics of the domain to which they are applied.

Download PDF sample

Adaptive Representations for Reinforcement Learning by Shimon Whiteson

by Paul

Rated 4.97 of 5 – based on 6 votes