ICML 2007

FIRST INTERNATIONAL WORKSHOP ON THE INDUCTION OF PROCESS MODELS

Several engineering and scientific tasks involve systems whose states change over time and space. Researchers model these systems not only for purposes of forecasting but also to identify the processes and interactions that drive the system's behavior. While a large literature on regression and time-series analysis exists, the models produced by these methods describe the shape of the trajectories but fail to capture the underlying processes. These methods focus on the predictive abilities of the models. They leave little room for incorporating domain knowledge and lead to models that make little contact with relevant scientific concepts. To address these and other limitations, researchers have explored methods that induce process-based models, which capture the complex temporal and spatial relationships generating the observed behavior.

Process-based models capture qualitative or quantitative aspects of the system behavior. Qualitative approaches include learning state transition models, Petri-nets, and biochemical reaction networks as well as learning from (time-stamped) event sequences. Quantitative approaches learn models grounded in standard mathematical representations (e.g., systems of differential equations). The challenge of crafting models and approaches that integrate qualitative and quantitative aspects of system behavior are also within the scope of process-based modeling. Methods for learning process-based models address problems important to machine learning such as knowledge-rich induction, learning in expressive languages and formalisms, and modeling temporal and spatial data.

In the ICML 2007 workshop on the induction of process models, we intend to bring scientists together and actively identify common research threads, define open problems, and develop collaborative contacts. It should provide a atmosphere more relaxed than in a conference setting where participants are encouraged to ask clarifying questions throughout the talks and to move past jargon-induced barriers. We aim to attract researchers with interests in learning dynamic causal models in several formalisms including Petri nets, qualitative and quantitative processes, differential equations, episode rules, logical rules, and others.

• learning Petri Net models
• learning using process algebra
• learning differential equation models
• learning in qualitative reasoning representations
• learning in temporal logic
• learning logical models of state transitions (e.g., by recursive clauses)
• learning from time-stamped event sequences (e.g., episode rules)
• learning from large databases of trajectories
• connectionist/subsymbolic models of sequence learning
• particularly welcome are case studies and applications, e.g., in engineering or the environmental, medical or biological sciences
• and papers identifying open problems such as dealing with missing data, noise handling, regularization, incorporating background knowledge, efficient search through the space of candidate process-based models, ...

Organizational Timeline

Abstracts (one page) should be submitted by April 27. Authors of accepted abstracts will be asked to submit a short 2 to 4 page paper in PDF format that describes their research in more detail. The papers will be placed on this website and made available to the public before the workshop takes place.

Important Dates

Format

Since we anticipate varied backgrounds of the participants, we will encourage speakers to present their work from a big-picture perspective and to clearly identify key issues in their research. After the presentations, we will lead an open discussion that begins with a summarization of the day's proceedings and themes in an attempt to identify key commonalities in research and immediate opportunities for collaboration. This discussion will also address the point of more frequent community interaction, covering topics such as sharing data and applications as well as the prospects of organizing a future meeting.

Program

Workshop Chairs

Will Bridewell, Stanford University, USA

Ljupco Todorovski, University of Ljubljana, Slovenia

Stefan Kramer, TU München, Germany

Organizing Committee

• W. Bridewell (Stanford Univ. , USA)
• S. Dzeroski (Jozef Stepan Institute, Slovenia)
• G.C. Garriga (Helsinki Univ. of Technology, Finland)
• S. Kramer (TU München, Germany)
• P. Langley (Stanford Univ., USA)
• H. Motoda (Osaka Univ., Japan)
• R. Otero (Univ. Coruna, Spain)
• B.J. Ross (Brock Univ., Canada)
• A. Srinivasan (IBM, India)
• L. Todorovski (Univ. Ljubljana, Slovenia)
• H. Toivonen (Univ. Helsinki, Finland)
• T. Washio (Osaka Univ., Japan)