"Applications of the Tensor Renormalization Group method to machine learning"

by Prof. Yannick Meurice (University of Iowa)

Monday 9 Oct 2017, 14:15 → 16:45 Europe/Berlin

3344 (TUM - Physics Department - Garching)

We briefly review the recently developed tensor renormalization group method for spin and gauge models. We discuss approximate solutions of the fixed point equation for the two-dimensional Ising model. We discuss the analysis of the images of stochastically generated loop configurations (“worms”) for this model near the critical temperature, by using RG adapted machine learning methods. We show numerical evidence supporting the conjecture that the leading eigenvalue of the Principal Component Analysis of the worm images has a logarithmic divergence at the critical temperature. We discuss the correspondence between the two approaches under coarse graining procedures.