Intelligent Pattern recognition and Applications
Patrick S. Wang Northeastern University
How do people learn and recognize things? These amazing capability has been taken for granted for years. It is until recently, when one tries to use computers or machines to do things like recognizing handwritten characters or faces, it becomes clear that such seemingly trvial tasks by human being turns out extremely difficult, if not impossible, by mechanical means such as computers. After decades of rigorous attacks, such research is still as fresh as ever, and such mystery as for how human beings can do it remains largely unknown but just began to unfold. In a sense, ``human brain" is the ``smartest" or the ``most intelligent" mechanism than any computer can provide. While no one knows exactly the detailed, sophisticated organism in human brains, the best way one can do, is perhaps to ``simulate" what ``might" be going on. In a way, the study of pattern recognition and artificial intelligence serves the purpose. We will get some inside views of pattern recognition techniques using artificial intelligence (AI) and neural network (NN) methodologies, and its applications to a lot of interesting and important problems including optical character recognition (OCR), zip-code recognition, bank check recognition, industrial parts inspection, scene analysis and image understanding, computer vision, and learning.
For the past few decades, there is a growing interest in the study of artififcial intelligence and rule-based expert systems. Pattern recognition plays an important role in such systems. In fact, there is now much interaction between expert systems and pattern analysis. It is interesting to see that the core of pattern recognition, in cluding ``learning techniques" and ``inference" also plays an important and central role in artificial intelligence. Visual perception, scene analysis, and image understanding are also essential to robotic vision. On the other hand, the methods in artificial intelligence such as knowledge representation, semantic networks, and heuristic searching algorithms can also be applied to improve the pattern representation and matching techniques in many pattern recognition problems -- leading to ``smart" pattern recognition. Moreover, the recognition and understanding of sensory data like speech or images, which are major concerns in pattern recognition, have always been considered as important subfields of artificial intelligence.This tutorial covers the following contents:
Dr. Patrick S. Wang is IAPR Fellow, tenured full professor of computer science at Northeastern University since 1983, research consultant at MIT Sloan School since 1989, and adjunct faculty of computer science at Harvard University Extension School since 1985. He received his Ph.D. in C.S. from Oregon State University, M.S. in I.C.S. from Georgia Institute of Technology, M.S.E.E. from National Taiwan University and B.S.E.E. from National Chiao Tung University. He was on the faculty at University of Oregon and Boston University, and senior researcher at Southern Bell, GTE Labs and Wang Labs prior to his present position. Dr. Wang was elected Otto-Von-Guericke Distinguished Guest Professor of Magdeburg University near Berlin, Germany, Fall 1996.
In addition to his research experience at MIT AI Lab, Prof. Wang has been visiting professor and invited to give lectures, do research and present papers in a number of countries from Europe, Asia and many universities and industries in the U.S.A. and Canada. Dr. Wang has published over 120 technical papers and 17 books in Pattern Recognition, A.I. and Imaging Technologies and has three OCR patents by US and Europe Patent Bureaus. As IEEE senior member, he has organized numerous international conferences and workshops and served as reviewer for many journals and NSF grant proposals. Prof. Wang is currently founding Editor-in-Charge of Int. J. of Pattern Recognition and A.I., and Editor-in-Chief of Machine Perception & Artificial Intelligence by World Scientific Publishing Co. and elected chair of IAPR-SSPR (Int. Asso. of P.R.). In addition to his technical achievements and contributions, Prof. Wang has been also very active in community service, and has written several articles on Verdi, Puccini, Bizet, and Wagner's operas, and Mozart, Beethoven, Schubert and Tchaikovsky's symphonies.
Linguistic Geometry: Winning Strategies for Multiagent Systems
|Speaker:||Boris StilmanUniversity of Colorado at Denver|
Linguistic Geometry (LG) is a mathematical model for knowledge representation and reasoning about large-scale multiagent systems. A number of such systems including air/space combat, robotic manufacturing, software re-engineering, Internet cyberwar, etc. can be modeled as abstract board games. These are multi-player games whose moves can be represented by means of moving abstract pieces over locations of an abstract board. The dimensions of the board (2D, nD, and even non-linear), its shape and size, the mobility of pieces, the turn of motions (including concurrent motions) - all can be tailored to model a variety of multiagent systems. Abstract board games are introduced here as a class of Complex Systems. The purpose of LG is to provide strategies to guide the participants of a game to reach their goals. Traditionally, finding such strategies required searches in giant search trees. Such searches are often beyond capabilities of modern and even conceivable future computers. LG dramatically reduces the size of the search trees, thus making the problems computationally tractable. LG relies on the formalization and abstraction of search heuristics of advanced experts. Essentially, these heuristics replace search by construction of strategies. The formalized expert strategies yield efficient algorithms for problem settings whose dimensions may be significantly greater than the ones for which the experts developed their strategies. Moreover, these formal strategies allowed to solve problems from different problem domains far beyond the areas envisioned by the experts. It is really fascinating that for certain classes of problems these expert strategies yield provably optimal solutions. Formalization employs formal linguistics, the theory formal languages. Since both, the formal languages and the geometry of heuristics, were involved, this approach was named Linguistic Geometry.
A number of prototypes of LG systems and commercial products were developed at Lockheed Martin Corp., GIS Solutions, Sandia National Laboratories, US Air Force Phillips Laboratory, Rockwell International Corp., University of Denver, and University of Colorado at Denver. These days a number of researchers and entrepreneurs are developing a framework for organizing an international multi-million dollar private VENTURE for developing industrial applications of LG.
The tutorial includes introduction to the subject and a brief research background. Basic definitions and a survey of the LG formal tools will follow. A number of experiments with LG tools will be described in details. They demonstrate how the LG tools solve problems of gradually increasing complexity. Besides interesting actual results, a detailed description of experiments is intended to develop an intuitive understanding of LG tools and their underlying structure. Armed with this intuition, a tutorial attendee will be able to digest a formal description of the LG tools given next. A new, much deeper account in the foundations of LG will be introduced. By redeveloping one of the experiments discussed earlier, it points a new direction in LG: solving search problems by construction of strategies (without any search at all). In the end we will discuss some issues of computational complexity. LG tools allowed to identify a wide subclass of tractable (polynomial) problems among those that are usually considered as intractable (exponential). The LG tools provide algorithms for their solution.
Professor Stilman received his M.S. degree in Mathematics in 1972 from Moscow State University, Moscow, USSR, and his two Ph.Ds, in Computer Science and Electrical Engineering, from the National Research Institute for Electrical Engineering, Moscow, in 1984. Currently, he is a Professor of Computer Science and Engineering at the University of Colorado at Denver, Denver, USA. He joined this school in 1991 as a tenure-track Associate Professor and was promoted to the rank of Full Professor with tenure in 1994. He spent a year (1990-91) as a Visiting Professor at McGill University, Montreal, Canada. His previous position of Chief of Department (1988-90) was with the Computer Division of the National Research Geological Institute for Oil Development, Moscow. Prior to joining that Institute, for sixteen years Dr. Stilman was with the National Research Institute for Electrical Engineering, Moscow. He has published over 150 papers on Artificial Intelligence and Software Engineering, and has written and contributed to the books published in the former USSR, USA, and Germany. His new book "Linguistic Geometry: From Search to Construction" has been published by Kluwer Acad. Publ. in 1999.