Learning two-dimensional spatial dynamics from experimental data

This thesis discusses the analysis of complex spatial dynamics using a computer learning algorithm. The goal is to model experimental data, the dendritic solidification of ammonium bromide crystals, using a learning algorithm to search through a space of possible models in order to find an optimal description of the data. The space of possible models is a class of probabilistic cellular automaton rules, a rule which is inherently local. The traditional definition of a cellular automaton has been enhanced here to include information which is non-local in both space and time thus allowing the models to reproduce a greater variety of complex spatial dynamics. The learning algorithm performing the stochastic search through the model space is a variation of the genetic algorithm. The technique is first applied to pattern data generated by deterministic models for the solidification process. Simple cellular automata and more complicated generalizations of cellular automata are used to generate test data for the learning algorithm. Video images of solidifying ammonium bromide dendrites are then modeled using the genetic algorithm, and the results are compared to the test cases.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Physics, General

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2011

Author dc:creator

• Richards, Fred Christian

Contributors dc:contributor

• Packard, Norman H.

• Physics, General
• Artificial Intelligence

Statement dc:rights

• Copyright 1991 Richards, Fred Christian

Language dc:language

eng

Identifier

AAI9124474
(UMI)AAI9124474

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/23576