# Theory Group Seminar

*,*Graduate Student

*,*

In this work, we study our ability to learn physical operations in quantum systems where every action is a priori unknown. We provide fundamental results for understanding what can be learned. As an application, we give a rigorous neural network algorithm for learning noisy quantum devices that provably work in many settings. For instance, assuming the noise on some non-(Clifford+T) gates are Pauli channels, our algorithm can learn the noisy states, gates, and measurements in an n-qubit quantum device to epsilon-error using O(log(n) / epsilon^2) experiments, even if the gate noise depends on other gates non-locally.

We numerically demonstrate the efficiency of our algorithm in learning all states, gates, and measurements in a 100-qubit device. This result contrasts with gate set tomography, which learns the relations between gates instead of the actual physical descriptions. In the second application, we show that when it is impossible to completely learn a noisy quantum device, we can still partially learn the device and use the device to store quantum data and perform quantum computation. We establish a large quantum advantage for algorithms using the noisy quantum device over any algorithm without quantum memory.