Description

Developing systems that learn how to perform complex tasks is a challenging task. As the knowledge to learn becomes complex, with diverse procedural constructs and uncertainties to be validated, the system needs to integrate a wider range of learning methods with different strengths. The Poirot system pursues such a multi-strategy learning methodology that employs multiple integrated learners and knowledge validation modules Burstein et al., 2007.

In such an integrated learning system, activities of participating agents have to be coordinated systematically, especially to ensure that the overall system acquires the desired procedural knowledge. This can be done through a meta reasoning capability that monitors and controls the learning process. The meta-level reasoner may solve a given learning problem with a set of meta-level plans for coordinating and prioritizing the activities of the agents based on their capabilities and on the status of learning.

Maven is a meta-level reasoner that coordinates activities of different learners in an integrated learning system. Maven is developed based on our early work on learning goals and meta-level reasoning for interactive knowledge capture Kim and Gil, IJHCS-2007 and Kim and Gil, AIEd-2003.

The system explicitly represents learning goals, a set of plans for achieving learning goals, and high-level learning strategies that prioritize learning goals. Maven includes both top-down and bottom-up control strategies, and therefore supports flexible interactions among learners. The status of desired learning goals and goal achievement history enables assessment of learning progress over time. Our current work focuses on developing a meta reasoner for Poirot Burstein et al., 2007.

More (Research)