Wolfram Science

Larry Werth: Associative Pattern Memory

Perception and Analysis lecture series

Larry Werth describes the construction and application of a system for associative pattern memory

Larry explores the basis for a computer-implementation of human-like associative pattern memory. Associative pattern memory systems have their origins in randomly connected neural network models, where terminal states end in cycles. Each state is randomly mapped to an input pattern, and the sampled input along with the current state determine the next state.

Using cycles allow for the creation of fuzzy hashing, as well as being a simple and fast implementation. It also yields a common language for pattern types, and integrates spatial and temporal information to create higher-level input patterns.

These techniques have several applications: hand-printed character recognition, machine vision, video compression, financial pattern forecasting and vector quantization in signal processing.

Larry will be releasing his software library as a general purpose tool for pattern recognition development.

Samuel Chen: Achieve Computer Vision with 2D CA--An Introduction

Perception and Analysis lecture series

Samuel Chen from Harvard University discusses computer vision techniques with 2D CA

Samuel uses 2D CA as a direct filter on visual images, and identified a class of 2D CA with 5 neighbors and radius 1 that can identify horizontal, vertical and diagonal, 45-degree lines. This class is divided into types that perform particular operations: they highlight vertical or horizontal lines only, or only diagonal lines. Other types multiply (think "thicken") horizontal or diagonal lines  in various ways. Additionally, various effects on other diagonal lines and small curved lines segments were cataloged.

One immediate application for the filters identified is to help differentiate between signal and noise in a time series of images, for example.