Experiments in Transfer Across Multiple Learning Mechanisms (2006)

by Nicholas A. Gorski. John E. Laird.

Proceedings of the ICML-06 Workshop on Structural Knowledge Transfer for Machine Learning

We present three different learning mechanisms for transferring spatial knowledge from one problem to another, related problem: memory-based transfer, search-based transfer, and transfer using reinforcement learning. We describe the approaches and present preliminary results dem-onstrating successful transfer using these ap-proaches in Soar and testing them in the Urban Combat Testbed. The study of transfer in learning is typically restricted to variants of a single learning mechanism. When multiple learning mechanisms are studied, underlying reasons for any differences in performance can be unclear, obscured by differences in task knowledge and the underlying per-formance system. In this work, we study three different learning mechanisms, all within the same general cogni-tive architecture (Soar) and all performing the same tasks, minimizing differences outside of the learning systems. The three learning mechanisms that we investigated are: storing map knowledge in short-term working memory; using chunking to learn operator selection rules; and using reinforcement learning to tune operator selection rules. [via]

PDF Finder

Experiments in Transfer Across Multiple Learning Mechanisms (2006)

Related Files

Rate this Document

Popular eBooks

ADS

Tag Clouds

Last Download

BookShelf