I’m starting my evolutionary programming again after a 2+ year hiatus. I decided to change my approach somewhat. Some of you may remember that my EP swarm I built was referred to as ‘my bees’. However, they were somewhat suicidal for reasons that I still do not understand, and the results they gave were useless.
My new approach will still use the ‘swarming’ technique, but this time the goal of the system will be to optimize an evolutionary computing neural network. This makes the ‘hive mindset’ I was programming before a bit more structured and less chaotic. Because of this structure, and the system is now truly a data-mining application, I decided to call my new implementation of the ‘swarming’ technique ‘Migo’ after the Lovecraftian Migo.
In this system, each computational node can evolve and the neural pathways can be taught. In the bees approach, each program evolved and acted alone. The ‘hive’ was constantly changing in fairly random ways and never converged correctly. In the Migo approach, the neural network is the end goal and the number of components that can evolved are smaller, which should be easier to see if its working and what changes to make.
I’ve always enjoyed my bees and I will try to get them working again one day. Though I’m hopefully the Migo experiment will have better results.