“Neural Evolution” (this image was created with the assistance of DALL·E 2)

Are you curious about the possibility of evolving artificial neural networks, akin to how biological neural networks have evolved to become more complex over time? In this blog post, I’ll walk you through my implementation of NeuroEvolution of Augmenting Topologies (NEAT), a method proposed by Stanley and Miikkulainen that uses evolutionary principles to guide hyperparameter optimization in neural networks.

If you’re a fan of the Battletech universe, you might know that the whole thing started from a tabletop game - the sort where you have miniatures of mechs on a map composed of hexagonal terrain tiles. However, the game can get quite expensive (buy miniatures) as well as time consuming (complicated rules), which both make the game hard to enjoy for casual players. So, why not create your own Battletech-inspired game that is both affordable and easy to play?

“Multiverse Kitchen Table” (this image was created with the assistance of DALL·E 2)

I started this project because I was curious about a sub-field of machine learning which is currently more hyped than quantum computing. Companies spend millions on this technology just because it is so versatile and applicable in our everyday lives. Three words: Large Language Models.