Neural Network Field Theory

Neural Network Field Theory (NNFT) explores the deep connections between the statistical mechanics of neural networks and quantum field theory, aiming to leverage tools and insights from one field to advance the other. Current research focuses on establishing rigorous mappings between neural network architectures and field theories, particularly investigating how training dynamics correspond to renormalization group flows and how to construct neural networks that realize specific field theories. This interdisciplinary approach offers new theoretical frameworks for understanding neural network behavior, potentially leading to improved training algorithms and more efficient network architectures for various applications, including 3D scene representation and human model fitting.