An experimental evolutionist machine learning.
Last year, I learned a bit about machine learning, especially the learning process of neural networks.
In the classic approach of machine learning, we measure a success score comparing the computed output and the expected one, and use it to improve the network. But this implies there is a "good" answer, and that we know it.
For some time now, I have wondered if we could apply the concept of species evolution to machine learning : define a neural network as a "specie", implement a mutation process, survival conditions and then simply wait for the survival of the most adapted version.
The aim of this project is to try and do that.