Representations for Learning Action Selection from Real-Time Observation of Task Experts

Mark A Wood, Joanna J Bryson

Research output: Contribution to conferencePaper

Abstract

The association of perception and action is key to learning by observation in general, and to program-level task imitation in particular. The question is how to structure this information such that learning is tractable for resource-bounded agents. By introducing a combination of symbolic representation with Bayesian reasoning, we demonstrate both theoretical and empirical improvements to a general-purpose imitation system originally based on a model of infant social learning. We also show how prior task knowledge and selective attention can be rigorously incorporated via loss matrices and Automatic Relevance Determination respectively.

Conference

Conference20th International Joint Conference on Artificial Intelligence
CountryIndia
CityHyderabad
Period6/01/0712/01/07

Cite this

Wood, M. A., & Bryson, J. J. (2007). Representations for Learning Action Selection from Real-Time Observation of Task Experts. 641-646. Paper presented at 20th International Joint Conference on Artificial Intelligence, Hyderabad, India.

Representations for Learning Action Selection from Real-Time Observation of Task Experts. / Wood, Mark A; Bryson, Joanna J.

2007. 641-646 Paper presented at 20th International Joint Conference on Artificial Intelligence, Hyderabad, India.

Research output: Contribution to conferencePaper

Wood, MA & Bryson, JJ 2007, 'Representations for Learning Action Selection from Real-Time Observation of Task Experts' Paper presented at 20th International Joint Conference on Artificial Intelligence, Hyderabad, India, 6/01/07 - 12/01/07, pp. 641-646.
Wood MA, Bryson JJ. Representations for Learning Action Selection from Real-Time Observation of Task Experts. 2007. Paper presented at 20th International Joint Conference on Artificial Intelligence, Hyderabad, India.
Wood, Mark A ; Bryson, Joanna J. / Representations for Learning Action Selection from Real-Time Observation of Task Experts. Paper presented at 20th International Joint Conference on Artificial Intelligence, Hyderabad, India.6 p.
@conference{73bb8c25475a4b86aa46efbe85fc0783,
title = "Representations for Learning Action Selection from Real-Time Observation of Task Experts",
abstract = "The association of perception and action is key to learning by observation in general, and to program-level task imitation in particular. The question is how to structure this information such that learning is tractable for resource-bounded agents. By introducing a combination of symbolic representation with Bayesian reasoning, we demonstrate both theoretical and empirical improvements to a general-purpose imitation system originally based on a model of infant social learning. We also show how prior task knowledge and selective attention can be rigorously incorporated via loss matrices and Automatic Relevance Determination respectively.",
author = "Wood, {Mark A} and Bryson, {Joanna J}",
year = "2007",
language = "English",
pages = "641--646",
note = "20th International Joint Conference on Artificial Intelligence ; Conference date: 06-01-2007 Through 12-01-2007",

}

TY - CONF

T1 - Representations for Learning Action Selection from Real-Time Observation of Task Experts

AU - Wood,Mark A

AU - Bryson,Joanna J

PY - 2007

Y1 - 2007

N2 - The association of perception and action is key to learning by observation in general, and to program-level task imitation in particular. The question is how to structure this information such that learning is tractable for resource-bounded agents. By introducing a combination of symbolic representation with Bayesian reasoning, we demonstrate both theoretical and empirical improvements to a general-purpose imitation system originally based on a model of infant social learning. We also show how prior task knowledge and selective attention can be rigorously incorporated via loss matrices and Automatic Relevance Determination respectively.

AB - The association of perception and action is key to learning by observation in general, and to program-level task imitation in particular. The question is how to structure this information such that learning is tractable for resource-bounded agents. By introducing a combination of symbolic representation with Bayesian reasoning, we demonstrate both theoretical and empirical improvements to a general-purpose imitation system originally based on a model of infant social learning. We also show how prior task knowledge and selective attention can be rigorously incorporated via loss matrices and Automatic Relevance Determination respectively.

M3 - Paper

SP - 641

EP - 646

ER -