The computational study of quantum systems presents complex challenges not unlike those encountered in common machine learning applications such as image or speech recognition. Therefore, it is not surprising that paradigms for simulating condensed matter physics, quantum chemistry, or models of quantum computing are being upended by rapid developments in machine learning algorithms. As an example, I will demonstrate how feed-forward and convolutional neural networks can be repurposed for classification of quantum phases of matter through supervised learning, by training on "images" of atomic configurations. The success of these algorithms has motivated physicists to explore the use of stochastic neural networks, including modified RBMs, for the efficient compression and sampling of quantum wavefunctions. Finally, I will comment on the prospect for the future speedup of machine learning algorithms using quantum hardware.

Biography:

Roger Melko is an associate faculty at the University of Waterloo, associate faculty at the Perimeter Institute for Theoretical Physics, and the Canada Research Chair in Computational Quantum Many-Body Physics. He received his PhD from the University of California Santa Barbara in 2005, and spent two years as a Wigner Fellow at Oak Ridge National Laboratory. His research involves the development of computer strategies for the theoretical study of quantum materials, atomic matter, and quantum information systems. He was the recipient of the 2012 IUPAP Young Scientist Prize in Computational Physics, and the 2016 Canadian Association of Physicists Herzberg Medal for achievement by an early-career physicist.