This talk presents an overview of the open source project "Leela Chess Zero" (Lc0), a new type of chess engine based on the AlphaZero paper released by DeepMind. Lc0 learns to play chess solely by playing against itself without any human knowledge or intervention. Starting from scratch with random moves, its play improves each iteration using a powerful combination of reinforcement learning and neural networks. We rely both on an active open community for the code and computing resources as many petaflops are required to train a solid neural network. Currently the engine is among the four best in the world.
Speakers: Folkert Huizinga