Shastri@CIS.UPENN.EDU (Lokendra Shastri) (03/28/89)
IJCAI-89 WORKSHOP CALL FOR PARTICIPATION CONNECTIONIST AI? Motivation and Agenda The focus of the workshop is to define critical issues that comprise the problem of systematic rule governed processes and connectionist architectures. The outcome of the workshop is to elaborate what the problem is and to motivate cross-talk between the connectionist and AI research communities. Numerous claims and counter claims have been made about the nature of connectionist models and how they relate to rule governed behavior. We feel that some researchers tend to oversimplify connectionism and underestimate what it has to offer. At the same time some others make very strong claims about connectionism and tend to underestimate the complexity of the AI problem and ignore insights obtained over years of research in AI and cognitive science. We also feel that some underlying problems in the discussions have never been raised. Through this workshop we hope to gain a better understanding of specific issues related to the integration of rules with connectionist processing approaches and to be able to more clearly specify critical problems that need to be addressed if a reconciliation between the approaches is warranted. Specific issues to be discussed Introductory Discussions - (Session I) 1. There are a number of variations on connectionism such as parallel distributed processing, localist or structured connectionist models, neural nets. What are the core aspects of connectionist models? 2. What is a rule? Aspects of rules to be addressed include - structure and representation of rules and control of rule-based processes. Reconciling rules with connectionism -- the alternatives? (Session II) 1. Is there a clash between rules and connectionist architectures? It is often asserted that connectionist models are "non-symbolic" or "sub-symbolic", and hence, fundamentally different from traditional AI approaches. Examine this claim? 2. Should connectionist architectures compute rules? If so, what kind of rules? If not, how does one reconcile the approach with rules as characteristics of performance? Can connectionism contribute to AI? (Session III) 1. It is claimed that connectionism just provides an interesting implementation paradigm. What is meant by "an implementation paradigm"? Can an implementation paradigm offer crucial insights into problems? 2. Evaluate the contributions made by recent work in Connectionism to central problems in AI such as representation, reasoning, and learning. Format Our aim is to gather around 25 experts from within mainstream AI as well as connectionism to discuss the above issues in depth. The workshop will consist of three 3-hour discussion sessions spread over one and a half days There will not be any presentations but only moderated discussions. Participation Participation in the workshop is by invitation only and is limited to 25 persons. Anyone who has published on issues directly related to the workshop may apply. Please submit one two page abstract outlining your position on one or more topics to be discussed and a list of your recent publications on any of these topics. The abstract should be in 12 point font (the size of this text) and double spaced. (References may extend beyond the two page limit.) Send three copies of your submission by APRIL 17, 1989 to: Lokendra Shastri Computer and Information Science University of Pennsylvania, Philadelphia, PA 19104. Organizers: Helen Gigley Lokendra Shastri Army Audiology and Speech Center Computer and Information Science Dept Walter Reed Army Medical Center University of Pennsylvania Washington, D.C. 20012 Philadelphia, PA l9l04 hgigley@note.nsf.gov shastri@cis.upenn.edu Alan Prince Psychology Department Brandeis University Waltham, MA 02254 prince@brandeis.bitnet