Structural Abstraction Experiments in Reinforcement Learning Export

@inproceedings{citeulike:1700281, abstract = {A challenge in applying reinforcement learning to large problems is how to manage the explosive increase in storage and time complexity. This is especially problematic in multi-agent systems, where the state space grows exponentially in the number of agents. Function approximation based on simple supervised learning is unlikely to scale to complex domains on its own, but structural abstraction that exploits system properties and problem representations shows more promise. In this paper, we investigate several classes of known abstractions: 1) symmetry, 2) decomposition into multiple agents, 3) hierarchical decomposition, and 4) sequential execution. We compare memory requirements, learning time, and solution quality empirically in two problem variations. Our results indicate that the most effective solutions come from combinations of structural abstractions, and encourage development of methods for automatic discovery in novel problem formulations.}, author = {Fitch, Robert and Hengst, Bernhard and Suc, Dorian and Calbert, Greg and Scholz, Jason}, booktitle = {The 18th Australian Joint Conference on Artificial Intelligence}, citeulike-article-id = {1700281}, citeulike-linkout-0 = {http://www.cse.unsw.edu.au/~bernhardh/LEARAbstractions_AI05.pdf}, keywords = {abstraction, concurrent, hierarchical, knowledge, survey}, posted-at = {2007-09-27 14:50:23}, priority = {4}, title = {Structural Abstraction Experiments in Reinforcement Learning}, url = {http://www.cse.unsw.edu.au/~bernhardh/LEARAbstractions_AI05.pdf}, year = {2005} }

TY - CONF ID - citeulike:1700281 L3 - citeulike-article-id:1700281 N2 - A challenge in applying reinforcement learning to large problems is how to manage the explosive increase in storage and time complexity. This is especially problematic in multi-agent systems, where the state space grows exponentially in the number of agents. Function approximation based on simple supervised learning is unlikely to scale to complex domains on its own, but structural abstraction that exploits system properties and problem representations shows more promise. In this paper, we investigate several classes of known abstractions: 1) symmetry, 2) decomposition into multiple agents, 3) hierarchical decomposition, and 4) sequential execution. We compare memory requirements, learning time, and solution quality empirically in two problem variations. Our results indicate that the most effective solutions come from combinations of structural abstractions, and encourage development of methods for automatic discovery in novel problem formulations. TI - Structural Abstraction Experiments in Reinforcement Learning T2 - The 18th Australian Joint Conference on Artificial Intelligence KW - abstraction KW - concurrent KW - hierarchical KW - knowledge KW - survey AU - Fitch, Robert AU - Hengst, Bernhard AU - Suc, Dorian AU - Calbert, Greg AU - Scholz, Jason PY - 2005/// UR - http://www.cse.unsw.edu.au/~bernhardh/LEARAbstractions_AI05.pdf ER -