January 2000: First Page of Thinking that Leads to NEAT
The idea for NEAT was mostly formed in a single sitting in January of 2000. I wrote these four pages of notes in the process of thinking of it.
On this page, I set out to think of a neuroevolution algorithm that would somehow be able to address the competing conventions (differing topologies) problem while still evolving variable topology neural networks. Most of the work on this page is an attempt to discover a way to describe network structures that is convenient. I end up settling on a kind of "slot"-based encoding where each connection is a slot between brackets. I have not yet discovered NEAT but I am thinking hard by the end about how to know which slot overlaps with which.
Link to Page 2