Statistical properties of cellular automata in the context of learning and recognition: Part 2, inverting local structure theory equations to find cellular automata with specified properties

Conference · Fri Nov 18 04:00:00 EST 1988

OSTI ID:6432188

Gutowitz, H A

This is the second of two lectures. In the first lecture the map from a cellular automaton to a sequence of analytical approximations called the local structure theory was described. In this lecture the inverse map from approximation to the class of cellular automata approximated is constructed. The key matter is formatting the local structure theory equations in terms of block probability estimates weighted by coefficients. The inverse mapping relies on this format. Each possible assignment of values to the coefficients defines a class of automata with related statistical properties. It is suggested that these coefficients serve to smoothly parameterize the space of cellular automata. By varying the values of the parameters a cellular automaton network may be designed so that it has a specified invariant measure. If an invariant measure is considered a ''memory'' of the network, then this variation of parameters to specify the invariant measure must be considered ''learning.'' It is important to note that in this view learning is not the storage of patterns in a network, but rather the tailoring of the dynamics of a network. 7 figs.

• Research Organization: Los Alamos National Lab., NM (USA)
• DOE Contract Number: W-7405-ENG-36
• OSTI ID: 6432188
• Report Number(s): LA-UR-88-3927; CONF-8809268-1; ON: DE89005447
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990230* -- Mathematics & Mathematical Models-- (1987-1989)
AUTOMATION
COMPUTER NETWORKS
LEARNING
MATHEMATICAL MODELS
MEAN-FIELD THEORY
MONTE CARLO METHOD
NONLINEAR PROBLEMS
PATTERN RECOGNITION
PROBABILITY
STATISTICAL MODELS