ANNEvolve :: Evolution of Artificial Neural Networks

The 4Play project produced an ANN that could store information while solving a problem, and make use of that information. This is the type of activity usually performed by a computer program. It is therefore natural to compare these two alternate approaches. One of the points of comparison is the amount of RAM required. The 4Play ANNs have the equivalent of 1 bit of RAM per neuron, and we never successfully got any lower than 34 neurons. Many people were surprised that so many neurons were required, since the number of weights and biases was well over 1000. However, it appears that the neuron requirement is strongly dependent on the RAM requirement. The RAM requirement is greater than that needed just to store the information of the problem. In the case of a computer program, RAM is needed for scratchpad memory. In the case of the ANN, we don't exactly know what the additional RAM does, but one can hypothesize that it is equivalent to scratchpad memory. Mr. Richard's article is directed to determining how much RAM a computer program requires to solve the 4Play problem. This is likely to represent a lower limit for the number of neurons that a successful ANN could have. [mt]

The following articles are in PDF formatting.

4-Play Procedure Solution - Psuedocode

4-Play Procedure Solution

4-Play Procedural Game Analysis

Shakespeare, Darwin, and the Monkeys

Author: Mitchell Timin
Posted: 12/26/2003

How Simulated Evolution Works

Author: Mitchell Timin
Posted: 11/16/2003

Meet Annevolve's skydiving, mouseball collecting Unix Admin

Author: Eric Anderson
Posted: 11/14/2003

Species Learning and a Hypothesis About Brain Learning

Author: Mitchell Timin
Posted: 10/02/2003