Machine Intelligence with Neural Networks, Decision Trees and Support Vector Machines

Sunday, 1:30 - 4:30 p.m.

(Pavilion Suite 1)

 Dr. Okan Ersoy, Purdue University

Overview: In the last decade, machine learning has come of age in a number of directions among the most important applications for machine learning are classification and recognition, prediction, and data mining. The most important emerging computational techniques for this purpose are neural networks, decision trees and support vector machines.

Classification and recognition are very significant in a lot of domains such as multimedia, radar, sonar, optical character recognition, speech recognition, vision, agriculture and medicine. We will discuss how these methodologies compete and interact with each other to achieve best results A number of practical examples from real-world problems will be illustrated.

Prediction is an application domain of classical significance. Who would not like to predict stock market prices in the coming week? What types of signals are predictable?

How do linear versus nonlinear prediction techniques compare? What are the best techniques for prediction? We will discuss answers to such significant and practical questions from the point of view neural networks and support vector machines. Again, a number of real-world problems will be illustrated.

Data mining is streamlining the transformation of masses of information into meaningful knowledge. It is a process that helps identify new opportunities by finding fundamental truths in apparently random data. The patterns revealed can shed light on application problems and assist in more useful, proactive decision making. Typical techniques for data mining involve decision trees, neural networks, nearest neighbor clustering, fuzzy logic and genetic algorithms. These will be compared, and illustrated with a number of examples.
 

Instructor’s Background:  Okan K. Ersoy was born in Istanbul, Turkey, on September 5, 1945. He received the B.S.E.E. degree from Robert College in 1967, and the M.S. Certificate of Engineering, second M.S., and Ph.D. degrees from the University of California, Los Angeles, in 1968, 1971, and 1972, respectively. He was a Teaching and research Assistant in the Department of Electrical Sciences and Engineering, UCLA (1968-1972), Assistant Professor in the Department of Electrical Engineering, Bosphorous University (1972-1973), and Associate Professor in the second semesters at the same university (1976-1980). He joined the Center for Industrial Research, Oslo, Norway, as a Researcher in the Computer Science Division in 1973. He was a Visiting Scientist at UCSD in 1980-1981. He has been with Purdue University, School of Electrical and Computer Engineering, West Lafayette, Indiana, Since August 1985. He was a Fullbright Fellow in 1976-78. His current interests include neural networks and applications, Fourier-related transforms, digital signal/image processing and recognition, fast and parallel algorithms and architectures, optical information processing and diffractive optics. He has published approximately 160 papers in his areas of interest. He published a book entitled Fourier-Related Transforms, Fast Algorithms and Applications (Prentice Hall, 1996). He also holds three patents that are separately patented in USA, Norway, Denmark, and Sweden. He has been a referee for IEEE Tran. Acoustics, Speech, Signal Processing, IEEE Tran. Computer, Applied Optics, Optical Engineering, a number of other journals, and granting agencies such as the National Science Foundation. He is associate editor of IEEE Tran. Neural Networks, IEEE Tran. Circuits Systems, and International Journal of Smart Engineering Systems Design. He has worked with numerous projects in his areas of interest in USA, Norway, and Turkey.