Deep Neural Networks Framed as Lattice Gauge Theories in Physics Research

Researchers have developed a novel framework that conceptualizes deep neural networks as lattice gauge theories, drawing parallels to high energy physics. This approach modifies existing NN/QFT duality to incorporate the layerwise permutation symmetry of neural networks. In this model, each layer of neurons functions as a lattice site, and weight matrices act as gauge fields. The study details the computation of the neuron-neuron propagator and introduces Feynman diagram machinery for perturbative analysis in 1/N, offering a theoretical basis for understanding information propagation in deep networks. AI

IMPACT This theoretical framework could offer new perspectives for understanding the internal dynamics and information propagation within deep learning models.

RANK_REASON Academic paper detailing a theoretical framework connecting deep neural networks to lattice gauge theories. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Deep Neural Networks
• Feynman diagram
• high energy physics
• Lattice gauge theories
• NN/QFT duality
• quantum field theory

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →