Rigor with Machine Learning from Field Theory to the Poincaré Conjecture

Sergei Gukov,James Halverson,Fabian Ruehle

2024-02-21

Abstract:Machine learning techniques are increasingly powerful, leading to many breakthroughs in the natural sciences, but they are often stochastic, error-prone, and blackbox. How, then, should they be utilized in fields such as theoretical physics and pure mathematics that place a premium on rigor and understanding? In this Perspective we discuss techniques for obtaining rigor in the natural sciences with machine learning. Non-rigorous methods may lead to rigorous results via conjecture generation or verification by reinforcement learning. We survey applications of these techniques-for-rigor ranging from string theory to the smooth $4$d Poincaré conjecture in low-dimensional topology. One can also imagine building direct bridges between machine learning theory and either mathematics or theoretical physics. As examples, we describe a new approach to field theory motivated by neural network theory, and a theory of Riemannian metric flows induced by neural network gradient descent, which encompasses Perelman's formulation of the Ricci flow that was utilized to resolve the $3$d Poincaré conjecture.

High Energy Physics - Theory,Machine Learning

What problem does this paper attempt to address?

This paper discusses how to use machine learning methods to obtain rigorous results in fields such as theoretical physics and pure mathematics that value rigor and understanding. The paper proposes two main approaches: one is to generate conjectures using machine learning and have them verified by human experts, and the other is to ensure the rigor of the results from the beginning using theoretical machine learning. Specific applications include using neural networks and reinforcement learning in areas such as string theory, algebraic geometry, lattice theory, and the Poincaré conjecture to discover new theorems or solve mathematical problems. In addition, the paper also introduces new approaches of applying neural network theory to field theory and metric flow, such as the correspondence between neural networks and quantum field theory and the Ricci flow induced by neural networks.

Rigor with Machine Learning from Field Theory to the Poincaré Conjecture

Mathematical Results Inspired by Physics

Machine learning in physics and geometry

Graph Laplacians, Riemannian Manifolds and Their Machine-Learning

A perspective on machine learning and data science for strongly correlated electron problems

Exploring the Truth and Beauty of Theory Landscapes with Machine Learning

Machine learning of discrete field theories with guaranteed convergence and uncertainty quantification

Decoding Quantum Field Theory with Machine Learning

Machine learning with quantum field theories

Perelman's proof of the Poincaré conjecture: a nonlinear PDE perspective

Machine Learning and Computational Mathematics

From the String Landscape to the Mathematical Landscape: a Machine-Learning Outlook

Robust learning from noisy, incomplete, high-dimensional experimental data via physically constrained symbolic regression

Machine Learning meets Number Theory: The Data Science of Birch-Swinnerton-Dyer

Learning Curves for Deep Neural Networks: A Gaussian Field Theory Perspective

Advancing mathematics by guiding human intuition with AI

Machine Learning Memory Kernels as Closure for Non-Markovian Stochastic Processes

Machine Learning Topological States

Importance of the Mathematical Foundations of Machine Learning Methods for Scientific and Engineering Applications

Revolutionizing physics: a comprehensive survey of machine learning applications