Machine learning of quantum many-fermion systems

The application of state-of-the-art machine learning (ML) techniques to statistical physics problems has seen a surge of interest for their ability to discriminate phases of matter by extracting essential features in the many-body wavefunction or the ensemble of correlators sampled in Monte Carlo simulations.

In this talk, I will focus on quantum many-fermion problems and demonstrate that convolutional neural networks (CNNs) can identify a plethora of collective states including metals, spin-density and charge-density wave ordered phases as well non-trivial states such as superconductors and topologically ordered states. Both supervised and unsupervised ML approaches will be introduced. I will further elucidate how CNNs can also be used to alleviate the notorious sign problem in fermionic quantum Monte Carlo techniques.

Joint work with Peter Broecker.

[1] P. Broecker et al., Scientific Reports 7, 8823 (2017)
[2] P. Broecker et al., arXiv:1707.00663