I spent most of yesterday working on the neural net project with Philip.

We're still in the process of working through the kinks in the basic NEAT engine, but we're making progress. It's a little slow, but our implementation has lots of moving parts and quite a bit of complicated stuff going on.

Each iteration, we're basically:

1) Generating a list of chromosomes which each encode for an artificial neural net
2) Transcribing those chromosomes into a population of artificial neural nets
3) Inputting data into each of those neural nets and seeing how they perform, then returning a fitness value
4) Eliminating a particular percentage of the population (e.g. 80%) with lower fitness values
5) Creating new offspring, some of which are mutated clones of the survivors, others which are created using a genetic crossover mechanism
6) And then doing the whole thing all over again for as many generations as we want

Meanwhile, we're trying to keep as much historical data as possible. Some is need for the purposes of crossover, and some is desirable for diagnostic purposes (i.e., it's useful to see how a particular lineage evolved, which individuals died off at what part in the process, etc.).

There are aspects of NEAT we haven't even implemented yet, such as speciation, which subdivides the population into subpopulations in order to protect innovative mutations for longer periods.

But as I say, we're making progress. Soon we'll have our basic tests completed and we'll be able to move on to game domains.

At that point I'm toying with the idea of setting up a project website, perhaps with an area where you can play Tic-Tac-Toe or Pente against our neural net AIs.

Until then, you'll just have to settle for the occasional blog update.