Boltzmann machines linked to Feynman path integrals

A new paper on arXiv (2301.06217) establishes a formal equivalence between Boltzmann machines used in machine learning and Feynman path integrals from quantum statistical mechanics. The authors show that hidden layers in neural networks correspond to discrete path elements in the path-integral formalism. This connection suggests that machine learning networks accumulate path-weights to map inputs to outputs, analogous to quantum interference. As a direct result, the paper provides general quantum circuit models applicable to such networks.