ResearchSpace

Feature selection for domain knowledge representation through multitask learning

Show simple item record

dc.contributor.author Rosman, Benjamin S
dc.date.accessioned 2014-12-22T07:38:13Z
dc.date.available 2014-12-22T07:38:13Z
dc.date.issued 2014-10
dc.identifier.citation Rosman, B.S. Feature Selection for Domain Knowledge Representation through Multitask Learning. IEEE International Conference on Development and Learning and on Epigenetic Robotics (ICDL-EpiRob), Genoa, Italy, 13-16 October 2014. en_US
dc.identifier.uri http://www.benjaminrosman.com/papers/icdl14.pdf
dc.identifier.uri http://hdl.handle.net/10204/7825
dc.description IEEE International Conference on Development and Learning and on Epigenetic Robotics (ICDL-EpiRob), Genoa, Italy, 13-16 October 2014. Abstract only version. en_US
dc.description.abstract Representation learning is a difficult and important problem for autonomous agents. This paper presents an approach to automatic feature selection for a long-lived learning agent, which tackles the trade-off between a sparse feature set which cannot represent stimuli of interest, and rich feature sets which increase the dimensionality of the space and thus the difficulty of the learning problem. We focus on a multitask reinforcement learning setting, where the agent is learning domain knowledge in the form of behavioural invariances as action distributions which are independent of task specifications. Examining the change in entropy that occurs in these distributions after marginalising features provides an indicator of the importance of each feature. Interleaving this with policy learning yields an algorithm for automatically selecting features during online operation. We present experimental results in a simulated mobile manipulation environment which demonstrates the benefit of our approach. en_US
dc.language.iso en en_US
dc.publisher IEEE en_US
dc.relation.ispartofseries Workflow;13443
dc.subject Artificial intelligence en_US
dc.subject Representation learning en_US
dc.subject Multitask learning en_US
dc.subject Domain knowledge representation en_US
dc.title Feature selection for domain knowledge representation through multitask learning en_US
dc.type Conference Presentation en_US
dc.identifier.apacitation Rosman, B. S. (2014). Feature selection for domain knowledge representation through multitask learning. IEEE. http://hdl.handle.net/10204/7825 en_ZA
dc.identifier.chicagocitation Rosman, Benjamin S. "Feature selection for domain knowledge representation through multitask learning." (2014): http://hdl.handle.net/10204/7825 en_ZA
dc.identifier.vancouvercitation Rosman BS, Feature selection for domain knowledge representation through multitask learning; IEEE; 2014. http://hdl.handle.net/10204/7825 . en_ZA
dc.identifier.ris TY - Conference Presentation AU - Rosman, Benjamin S AB - Representation learning is a difficult and important problem for autonomous agents. This paper presents an approach to automatic feature selection for a long-lived learning agent, which tackles the trade-off between a sparse feature set which cannot represent stimuli of interest, and rich feature sets which increase the dimensionality of the space and thus the difficulty of the learning problem. We focus on a multitask reinforcement learning setting, where the agent is learning domain knowledge in the form of behavioural invariances as action distributions which are independent of task specifications. Examining the change in entropy that occurs in these distributions after marginalising features provides an indicator of the importance of each feature. Interleaving this with policy learning yields an algorithm for automatically selecting features during online operation. We present experimental results in a simulated mobile manipulation environment which demonstrates the benefit of our approach. DA - 2014-10 DB - ResearchSpace DP - CSIR KW - Artificial intelligence KW - Representation learning KW - Multitask learning KW - Domain knowledge representation LK - https://researchspace.csir.co.za PY - 2014 T1 - Feature selection for domain knowledge representation through multitask learning TI - Feature selection for domain knowledge representation through multitask learning UR - http://hdl.handle.net/10204/7825 ER - en_ZA


Files in this item

This item appears in the following Collection(s)

Show simple item record