dunc@eecg.toronto.edu (Duncan Elliott) (04/28/89)
Electrical Engineering Computer Group Cider Seminar Series A Neural Network for Temporal Pattern Recognition by Tet Yeap Electrical Engineering Computer Group University of Toronto Time: Tuesday, May 2, 1989, 12:05 --- Place: GB 220 Recent work on neural networks has concentrated on their ability to store, retrieve and process static patterns. However, in problems such as motion detection, speech processing and predictive signal processing, the information to be processed consists of a sequence of temporally related patterns. Neural networks that handle only static patterns are not suited for these problems. A neural network that can be trained to recognize a sequence of events in time is presented in this seminar. It utilizes a process of temporal integration, which is implemented by providing each node with a fading memory and the use of a special interconnect pattern. A simulation model for a motion detector based on the proposed network was implemented. It can be trained to distinguish between several speeds of a moving pattern of random dots. Coming Soon Date Who Topic May 19 Prof. Wortman, Adventures in Concurrent Compilation Michael Junkin -- Duncan Elliott, Dept. EE, University of Toronto, Toronto, Canada M5S 1A5 dunc@eecg.toronto.edu dunc@eecg.utoronto.ca uunet!utai!eecg!dunc LAT: 43 39' 35.9"N LON: 79 23' 41.7"W ELEVATION: 349.30 VE3PKD