What is this, and why should I care about it?

This is the result of about two years of work by one person, me, Joseph Reisinger. Basically it is a neural network that attempts to mimic some biological form (instead of an abstract representation) and in the attempt perform more complex computation efficiently.

What kind functions are you solving with this?

Right now there is only one simulation that provides input. Its a really simple "world" with a "creature" and lights and blocks. The creature gains "energy" when it is on a light and loses it otherwise. Very typical optimization problem. Obviously the goal of this project is not to solve only one problem, but rather a large variety, so the neural network has been abstracted from world simulation.

Thats all and good, but does it work?

More or less, yes. I have evolved something close to the simplest solution to the lightseeker problem. I say "close to" because the neural network is rather stochastic as a single DNA set (almost) never will grow the same pattern twice.

What work are you doing now? I've noticed that you haven't done much on this since like February.

I've been searching for ways to reduce the stochasticity of the network. Mainly by allowing each individual neuron to set more parameters based on evolution rather than hard setting them at the beginning. The more variables that the DNA controls the better evolved the outcome is. Another reason I haven't been working on alice lately is because I've been laying the groundwork for kozue (like Kozue from Utena), which is more of a non-classical neural net program. It will use more cellular autonoma-like methods to grow neural structures. Its cool.

Izumi is a cool idea that popped into my head. Its an interesting way of visualizing time-varying linguistic data (ie, websites and the like) and drawing connections between content. It will require some natural language processing (not to sure how in depth I will go with this part), so it should be pretty fun to work on. Oh, Izumi is from Katsura Masakazu's recent work I''s. It was published in Shounen Sunday Comics starting sometime in '99 I think; Izumi is the second foil love interest (so gee, how does that fit here...). Izumi is read: I as in "eek a mouse," zu again like zoo, mi is the name I call myself.

Izumi is limited because the text parser is very basic. For instance, "fish" and "fishes" are not considered similar, also "a lot" is a common word, but when broken up correctly becomes "a" and "lot" where "lot" is not considered common. This homonym misinterpretation can cause problems for very small dictionaries.

This project is hosted by SourceForge, they're really cool people.