MC 5501 and Zoom (email compmath@uwaterloo.ca for Zoom link)
Speaker
Professor Roger G. Melko | Department of Physics & Astronomy, University of Waterloo
Title
Generative models for quantum state reconstruction
Abstract
Generative models are powerful tools in unsupervised machine learning, where the goal is to learn an unknown probability distribution that underlies a data set. Recently, physicists have been adopting generative models from modern industry uses, such as computer vision or natural language processing (NLP), as tools for studying physical systems. In this talk, I will discuss how generative models are capable of reconstructing the state of a quantum computer, given projective measurement data from individual qubits. After training, these virtual models can be studied with probes unavailable to the original experiment, providing unique insight into the nature of the quantum state. I will outline the strategy for quantum state reconstruction using generative models and show examples on real data from a neutral atom quantum computer. I will discuss the continuing theoretical development of the field, including the exploration by physicists of powerful autoregressive models responsible for recent advances in modern NLP, such as recurrent neural networks and transformers.