Share Email Print

Proceedings Paper

Real-time modeling of primitive environments through wavelet sensors and Hebbian learning
Author(s): James M. Vaccaro; Paul S. Yaworsky
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Modeling the world through sensory input necessarily provides a unique perspective for the observer. Given a limited perspective, objects and events cannot always be encoded precisely but must involve crude, quick approximations to deal with sensory information in a real- time manner. As an example, when avoiding an oncoming car, a pedestrian needs to identify the fact that a car is approaching before ascertaining the model or color of the vehicle. In our methodology, we use wavelet-based sensors with self-organized learning to encode basic sensory information in real-time. The wavelet-based sensors provide necessary transformations while a rank-based Hebbian learning scheme encodes a self-organized environment through translation, scale and orientation invariant sensors. Such a self-organized environment is made possible by combining wavelet sets which are orthonormal, log-scale with linear orientation and have automatically generated membership functions. In earlier work we used Gabor wavelet filters, rank-based Hebbian learning and an exponential modulation function to encode textural information from images. Many different types of modulation are possible, but based on biological findings the exponential modulation function provided a good approximation of first spike coding of `integrate and fire' neurons. These types of Hebbian encoding schemes (e.g., exponential modulation, etc.) are useful for quick response and learning, provide several advantages over contemporary neural network learning approaches, and have been found to quantize data nonlinearly. By combining wavelets with Hebbian learning we can provide a real-time front-end for modeling an intelligent process, such as the autonomous control of agents in a simulated environment.

Paper Details

Date Published: 22 June 1999
PDF: 8 pages
Proc. SPIE 3696, Enabling Technology for Simulation Science III, (22 June 1999); doi: 10.1117/12.351188
Show Author Affiliations
James M. Vaccaro, Air Force Research Lab. (United States)
Paul S. Yaworsky, Air Force Research Lab. (United States)

Published in SPIE Proceedings Vol. 3696:
Enabling Technology for Simulation Science III
Alex F. Sisti, Editor(s)

© SPIE. Terms of Use
Back to Top