Abstract: Machine learning is making great strides in the natural sciences, yet its stochastic and opaque nature poses challenges for integration into rigorous disciplines like theoretical physics and pure mathematics. This colloquium explores strategies to achieve rigor in scientific research using ML. We highlight the use of applied ML to obtain rigorous results through conjecture generation and reinforcement learning, with applications ranging from string theory to knot theory and the smooth 4d Poincaré conjecture. We also propose direct connections between ML theory, theoretical physics, and pure mathematics. Illustrative examples include innovative approaches to field theory inspired by neural networks, and a theory of Riemannian metric flows derived from neural network gradient descent, extending Perelman's Ricci flow that was used to resolve the 3d Poincaré conjecture.
Join via Zoom:
Meeting ID: 818 6692 9019
The colloquium is held in Feynman Lecture Hall, 201 E. Bridge.