I would like to map the simplest 8X8 matrices, one to one, but am not sure which AI algorithm would give the best performance. I am thinking about the DeepLearning4j, however, I don't know which neural architecture to use.
I would like to make a simple and a very "stupid" bot for playing the chess. The result I am hoping to obtain is a system which can learn the chess rules rather than make intelligent moves (although it would be great if it could do that as well). I understand that chess bots are nothing new, however, I am not interested in making a chess bot but a system that can simply learn by giving nothing else than an 8X8 matrix as an input and 8X8 matrix as an output. Chess is irrelevant, and it can be any other game that can be represented with values within an 8X8 matrix.
I am aware of the newest image mappers that transform horses to zebraz, but I need something more precise and 1 to 1 learning.
The amount of data I can get for learning is also not an issue (I can generate as much as I want).