UCSC-CRL-94-22: MORPH II: A UNIVERSAL AGENT: PROGRESS REPORT AND PROPOSAL

06/01/1994 09:00 AM
Computer Science
This short report describes progress on the Morph Adaptive Pattern-Oriented chess system over the last four years and describes ongoing work on Morph II, a domain-independent machine learning system and problem-solver. Morph II, is a direct generalization of the original Morph model and incorporates significant improvements to all aspects of the original system. The Universal Agent (Morph II) falls directly within the hierarchical reinforcement learning paradigm that has been gaining popularity in the last few years, but goes beyond: in this learning model, the system is responsible for abstracting its own features and patterns, developing its own learning modules, and employing transformations on the representations used. In essence the system is learning neurally (through weight propagation), genetically (through pattern evolution) and symbolically (through its own abstractions). Still, the system maintains uniformity, in that each module at any level of the hierarchy is an instantiation of the same learning strategy. This report describes the motivations and philosophy behind MorphII, outlines current and future performance objectives, and discusses the potential impact of this new technology.

UCSC-CRL-94-22