IEEE's Conference on AI for Applications has been held 11 times since 1984. In February 1995, CAIA met in Los Angeles, across from the University of Southern California, 1 and included plenary speakers and panels addressing several critical issues in different areas:
- Knowledge discovery: Usama M. Fayyad, "A Machine Learning Approach to the Analysis of Large Image Databases;" Se June Hong, "Knowledge Discovery: Use of Contextual Information for Feature Ranking and Discretization;" and Mario Schkolnick, "Knowledge Discovery at IBM."
- Graphic models and uncertainty: Judea Pearl, "Graphical Models and Causality."
- New models of knowledge-based systems: Herb Schorr, "AI at ISI: New Models of Knowledge-Based Systems."
- Soft computing: Plenary Panel and Track—Bernadette Bouchon-Meunier (chair): Lofti Zadeh, I.R. Goodman, Abraham Kandel, Anca Ralescu, Enrique Ruspini, Ronald Yager, and John Yen. Editor Panel—Steve Cross ( IEEE Expert) and Ramesh Patil ( AI Magazine).
Recently, IEEE Expert published a short history of the conference. 2 Over the years, the conference has had a large participation throughout the world, with about 70% of the papers coming from the US and 30% from the rest of the world. Historically, participation has come from both universities and industry.
In 1995, conference participants presented 45 papers (of which 40 were published in the proceedings) and 12 poster papers (full papers were also published in the proceedings). There were 101 authors and coauthors of the 40 published and presented papers. Authors for roughly 90% of those papers came from academic organizations and 10% from industry, representing organizations from nine countries: Australia (9), Canada (6), France (8), Japan (10), Norway (2), Singapore (2), Taiwan (5), UK (6), and the US (55).
Of the 40 published papers, we are featuring five as a special AI Applications track in this and the current and the following two issues of Expert. Each received "outstanding" recommendations from its three reviewers.
- C.W. Liew's "Using Feedback to Improve VLSI Designs" describes a new technique called constrained redo that uses feedback to improve the power and coverage of an existing system for VLSI design. Typical knowledge-based expert systems are static, incorporating large amounts of domain-specific knowledge to generate good initial solutions. However, these systems cannot use information generated from analysis of the solutions to provide improved solutions. As a result, the authors use an approach that is part of the feedback-directed optimization framework to help generate improved designs.
- Chih-Hung Wu and Shie-Jue Lee's "Knowledge Verification with an Enhanced High-Level Petri Net Model" uses enhanced high-level Petri nets to model rule-based systems. After modeling the system as a Petri net, their approach uses markings to facilitate verification of the system by determining redundancy, subsumption, conflict, cycles, and unnecessary conditions.
- Jonathan Lee, Lein F. Lai, and Wei T. Huang's "Conceptual Graphs as a Basis for Expressing Task-Based Specifications" identifies several issues for mapping a task-based specifications methodology into conceptual graphs, including representation of constraints and state modeling, rigid versus soft post conditions, and the distinction and existence of different operators used to model specifications. In particular, the authors propose using conceptual graphs to express task specifications in which the specifications are driven by the task structure of the problem-solving knowledge. The approach permits verification of specification representations.
- Timothy Lenz, James K. McDowell, Martin C. Hawley, Ahmed Kamel, and Jon Sticklen's "A Decision Support Architecture for Polymer Composites Design: Implementations and Evolution," a multidisciplinary study, develops an architecture for a knowledge-based system for polymer composite material design. The authors use the generic task approach to develop an intelligent support system for polymer design. They trace the evolution of multiple systems designed to assist decision making, and contrast the different versions of the systems. Throughout the growth and development of the material design methodology, the overall vision for polymer composites design has evolved as well.
- James Mayfield, Marty Hall, and Tim Finin's "Using Automatic Memoization as a Software Engineering Tool in Real World AI Systems" describes how memoization can be made viable on a large scale. Memo functions and memoization refer to the tabulation of results of a set of calculations to avoid repeating the calculations. Automatic memoization refers to a method by which a function can be changed mechanically into one that memoizes or caches its results. The authors point out advantages and uses of automatic memoization not previously described and identify components of an automatic memoization facility.