How hard is it to learn a quantum state?
This event is part of the Condensed Matter Theory Seminar Series.
The fundamental difficulties in simulating quantum physics is one of the core motivations for building a quantum computer. As we enter the current era of NISQ hardware, we are faced with a new paradigm: devices that are difficult to simulate, but which can be measured to produce an abundance of data. At the same time, powerful machine learning methods are being adapted to learn representations of quantum states directly from measurement data. It is therefore fair to wonder whether difficulties in simulating large quantum systems also translate into difficulties in learning their states from data. In this talk I will use recent strategies for quantum state reconstruction based on generative modeling with neural networks to study the learnability scaling of prototypical groundstate wavefunctions in NISQ devices. The efficiency with which quantum states can be reconstructed in neural networks can be quantified numerically, and compared to expectations such as those given by tensor network theory. I will speculate on the implications that an answer to the question "how hard is it to learn a quantum state" would have, focusing on the current generation of experimental quantum simulators.