Chess as a Machine Learning Problem

I just learned about the minmax algorithms coupled with the Bayesian approach to penalize risks and choose the optimal decision in Machine Learning. It turns out that Stockfish uses the"Efficiently Updatable Neural Networks" (NNUE) approach which uses similar principles at its core. As a part of my ML project and being a chess fan I was thinking of trying a simpler version of chess, maybe with fewer pieces or a smaller size of the board, and build something that can converge to an actual mathematically perfect move, given a position.

I love the simplicity of the game which even a 3-year-old can understand and the infinite depths which can keep the players and researchers busy for their entire lives. Looking forward to trying some new things