All Posts

  • A Monte Carlo analysis of the board game Elder Sign

    Defeating Cthulu more effectively using probability theory.

    Read More
  • Measuring overfitting in multi-agent reinforcement learning

    While it is possible to use self-play to learn effective policies with little human input, agents can get stuck in local optima and overfit to opponent's policies.

    Read More
  • Learning to play snake at 1 million FPS

    Using advantage actor-critic to learn Snake in under 5 minutes on a massively parallel vectorised environment.

    Read More