Machine learning the effects of many quantum measurements

A novel aspect of recent experiments with quantum devices is that measurements can play an active role in preparing the state of the system, rather than just in diagnosing it. Unlike unitary evolution, the quantum collapse induced by local measurements can have a highly non-local impact on the state, instantaneously destroying or creating long range entanglement. I will begin by reviewing the surprising collective effects that can ensue, such as measurement induced phase transitions and new entanglement structures. There is, however, a fundamental barrier to observing such measurement induced phenomena, because the post-measurement state is conditioned on many-measurement outcomes with exponentially small probability of recurring. I will describe how we overcame this ‘post-selection problem’ by cross-correlating experimental data, taken with Google’s quantum processor, with results of a generative machine learning model trained on the experimental data. Our approach reveals that measurement-induced long range entanglement emerges where classical models lose the ability to learn quantum state properties, establishing a ‘learnability transition’ that marks a fundamental boundary in our ability to predict quantum behavior.

Speaker: Ehud Altman, UC Berkeley

Editor's Note: The original speaker, Mitch Begelman, University of Colorado, Boulder is unavailable.