Publikation

Using Joint Probability Densities for Simultaneous Learning of Forward and Inverse Models

Mark Edgington, Yohannes Kassahun, Frank Kirchner

In: Nils T. Siebel, Josef Pauli (Hrsg.). IEEE IROS International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS-09) 2nd International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot System October 11-15 St. Louis Missouri United States Seiten 19-22 10/2009.

Abstrakt

In this position paper we propose that in many cases, instead of using standard regression methods for directly capturing relationships between variables, joint probability density estimates can and should be used for this purpose. With a good joint probability density estimate, any relationship which exists between variables can be extracted in the form of a regression function. Depending on the chosen density estimate representation, a regression function can be derived with relatively little computational effort. In essence, this means that by learning a joint probability density, both forward and inverse models have been captured. This method of learning the relationships between variables is demonstrated through a series of experiments.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence