An Architecture and Language for the Integrated Learning of Demonstrations

Mark H. Burstein, Marshall Brinn, Mike Cox, Talib Hussain, Robert Laddaga, Drew McDermott, David McDonald, Ray Tomlinson

POIROT is an integration framework and reasoning control system that combines the products of a variety of machine learning mechanisms in order to learn and perform complex web services workflows, given a single demonstration example. POIROT's extensible multi-strategy learning approach to developing workflow knowledge is organized around a central hypothesis blackboard and representation language for sharing proposed task model generalizations. It learns hierarchical task models from semantic traces of user-generated service transaction sequences. POIROT's learners or hypothesis formers develop, as testable hypotheses, generalizations of these workflow traces by inferring task order dependencies, user goals, and the decision criteria for selecting or prioritizing subtasks and service parameters. Hypothesis evaluators, guided by POIROT's meta-control component, plan and execute experiments to confirm or disconfirm hypotheses extracted from these learning products. Hypotheses and analyses of hypotheses are represented on the blackboard in the language LTML, which builds on both OWL-S and PDDL.

Subjects: 12. Machine Learning and Discovery; 2. Architectures

Submitted: May 15, 2007

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.