Genetically-regulated neuromodulation facilitates multi-task reinforcement learning

Tools

Item Type:	Conference or Workshop Item
Title:	Genetically-regulated neuromodulation facilitates multi-task reinforcement learning
Creators Name:	Cussat-Blanc, S. and Harrington, K.I.S.
Abstract:	In this paper, we use a gene regulatory network (GRN) to regulate a reinforcement learning controller, the State- Action-Reward-State-Action (SARSA) algorithm. The GRN serves as a neuromodulator of SARSA's learning parame- ters: learning rate, discount factor, and memory depth. We have optimized GRNs with an evolutionary algorithm to regulate these parameters on specific problems but with no knowledge of problem structure. We show that genetically- regulated neuromodulation (GRNM) performs comparably or better than SARSA with fixed parameters. We then ex- tend the GRNM SARSA algorithm to multi-task problem generalization, and show that GRNs optimized on multi- ple problem domains can generalize to previously unknown problems with no further optimization.
Keywords:	Reinforcement Learning, Gene Regulatory Network, Parameter Control, Multi-Task Learning, Neuromodulation
Source:	Proceedings of the Conference on Genetic and Evolutionary Computation
Title of Book:	Proceedings of the 2015 on Genetic and Evolutionary Computation Conference - GECCO '15
ISBN:	978-1-4503-3472-3
Publisher:	Association for Computing Machinery
Page Range:	551-558
Date:	July 2015
Official Publication:	https://doi.org/10.1145/2739480.2754730

Repository Staff Only: item control page