Mathematics & Machine Learning Seminar

Google DeepMind is at the forefront of leveraging AI to push the boundaries of scientific discovery. This talk will provide an overview of two compelling projects that exemplify this approach: Reinforcement Learning for Nuclear Fusion and AlphaGeometry2. The RL for nuclear fusion project focuses on training an AI agent for real-time plasma control within a tokamak, showcasing the capability to generate and maintain a diverse array of plasma configurations with a simpler and versatile architecture compared to conventional control methodologies. AlphaGeometry2 demonstrates a powerful combination of language models and symbolic engines, achieving a groundbreaking milestone by surpassing gold medalist level performance on International Math Olympiad geometry problems for the first time.