The Game of Life is a cellular automaton whose evolution is determined by its initial state, requiring no further input. One interacts with the Game of Life by creating an initial configuration and observing how it evolves. This Kaggle challenge aims to explore whether machine learning algorithms can work backwards to the initial configuration.
With the game evolving step by step on a 25x25 square grid with, the challenge is similar to predicting a future frame from a video input. A starting approach is thus based on frame prediction. The initial algorithm first extracts single-step evolutions and trains a CNN to predict a new grid. Predicting earlier states is then done by running predicted states back through the CNN.
Preliminary results show this approach yields good performance and that the state of the initial configuration is a significant criteria of the capacity for the CNN to predict it. Specifically, a sparse grid appears to lead to the poorest results.
To see code and results, follow the link to my Github page.