Note: This workshop has been cancelled as of 4 April 2005.

ICML 2005 Workshop on
Knowledge-Intensive Learning in Simulated Task Environments

Description

This workshop, to take place on 7 August 2005 in Bonn, Germany, will focus on encouraging investigations of and discussions on knowledge-intensive approaches of machine learning (ML) techniques in the context of simulated task environments (e.g., computer games, artificial life, design testing, intelligent tutoring, medicine, multi-agent systems, music, space systems, military simulators). It will provide a forum for researchers who share interest in this topic and want to learn more about how techniques, used by others in diverse application domains, employ significant background knowledge to accomplish performance goals.

Rationale

The majority of ML research today focuses on knowledge-poor learning frameworks that process comparatively little background knowledge. Yet knowledge-intensive approaches are of particular interest because they support rapid learning, which is required by performance tasks that demand competent behavior after minimal experience. For this reason, this topic could quickly gain in popularity, especially as researchers increasingly focus on more challenging tasks.

Simulated task environments often define such tasks. For example, simulators for strategy games (e.g., turn-based, real-time, team sports), artificial life, medical, and computer-generated forces, among many others, usually involve interacting with complex processes, objects, and relations among them. These are frequently used for educational purposes to train skills and decision-making strategies. Importantly, simulators also provide researchers with data generation processes that can be used to conduct controlled experiments, and permit the study of learning techniques embedded in problem solving environments in which a premium is placed on rapid learning.

While researchers have investigated knowledge-intensive techniques, they characteristically have had to delve deeply into their application domain, which complicates sharing their insights and concerns with others who study related techniques on distinct domains (e.g., researchers who develop applications for space systems don't usually interact closely with those studying intelligent tutoring simulators). This workshop will provide a forum for them to speak with others who are addressing similar problems.

We expect this workshop to provide a snapshot on motivating tasks and existing techniques for knowledge-intensive learning. It should also increase awareness on and provide clear descriptions of shared/pertinent research issues that require further attention. This will assist in conducting formal empirical and analytic studies on algorithms and architectures that could extend the state-of-the-art.

Intended Audience

Knowledge-intensive learning and the use of simulations are germane to many ML sub-disciplines. However, researchers interested in rapid learning approaches, learning and planning, using ontologies in learning approaches, learning in complex computer games, simulation-based learning, delayed reinforcement learning in large decision spaces, and related topics should be particularly interested.

Topics relevant to this workshop include, but are not limited to, the following technical areas among a broad coverage of interesting application domains:

Learning to transfer knowledge among diverse simulation tasks
Learning via meta-level reasoning and representation
Learning and reasoning in dynamic physical worlds
Explanation methods for simulators
Knowledge-intensive behavioral cloning approaches
Learning techniques in training simulators
Adversary and adversarial strategy recognition and modeling
Learning and planning for large-scale environments
Multi-agent learning in simulation tasks
Social reasoning among virtual learning agents
Learning to model or simulate complex systems
Evaluation methodologies for knowledge-intensive learning
Analyses of knowledge intensive learning techniques
Relevant applications, lessons learned, and surveys

Format

We will begin with a short presentation of questions for framing the workshop that the committee wishes to address, along with some potential answers to encourage audience interaction. We will schedule invited presentations by leading researchers with differing perspectives. (At this time, Ivan Bratko and Tom Dietterich are planning to give these presentations, and we expect to invite others.) Substantial time will be reserved for pre-determined discussion topics (e.g., on existing and promising approaches that highlight and exemplify key research topics that require further study). We will invite authors of accepted submissions to present their work. We encourage demonstrations, and will reserve a time period for them as needed. We may form a panel of experts to address issues of interest that arise among the committee and submissions, and/or other activities to provide for a lively exchange of information among the participants.

For example, we invite participants to enter a games-related competition, using TIELT to access the simulator. (We are in the process of defining this competition, and will announce it at this site.) We will select performance tasks that require substantial background knowledge to obtain reasonable performance, and encourage the development, application, and comparison of learning approaches on these tasks. A time period will be reserved to describe the results of this competition, to include presentations of selected competitors.

Participation and Submissions

Participation is open to all ICML'05 attendees.

We seek submissions (max 6 pages in ICML'05 format, but please include the author names as this will not be a double-blind review process) that describe state-of-the-art work on this subject, compare existing approaches, and identify existing limitations of existing approaches. We also seek short text statements of interest from others who wish to attend. Please send all submissions and statements to David W. Aha.

Important Dates

1 April 2005	WS Paper submission deadline
22 April 2005	Notification of acceptance to submitters
13 May 2005	WS final paper deadline
20 May 2005	Workshop notes due (on-line)
7 August 2005 (CANCELLED!)	Workshop date (Bonn, Germany)

Workshop Committee

Agnar Aamodt, Norwegian University of Science and Technology
David W. Aha (co-chair), Naval Research Laboratory
Lawrence B. Holder, University of Texas at Arlington
Daniel G. Shapiro (co-chair), Applied Reactivity, Inc.
Gerhard Widmer, Johannes Kepler University