Authors: Daniel Hothem, Kevin Young, Tommie A. Catanatch, Timothy Proctor

 

Quantum computing is a leading candidate for achieving computational speedups in a post-Moore era. However, current devices are noisy and error-prone. In order for quantum computers to fulfill their promise, we need scalable methods for predicting the performance of a given device. As the noise exhibited by quantum devices can be quite complex, highly expressive models may be necessary to accurately predict the behavior of a quantum device. In this work, we explore the potential of neural networks to play this role. We demonstrate a convolutional neural network’s ability to learn both simulated and experimental error models; in each case outperforming non-neural network models that are based on per gate error rates. We also investigate how a convolutional network’s ability to learn error models varies as a function of dataset size and measurement accuracy.

 

This work was supported by the LDRD program at Sandia National Labs. Sandia National Labs is a multimission laboratory managed and operated by NTESS, LLC, a wholly owned subsidiary of Honeywell International Inc., for DOE’s NNSA under contract DE-NA0003525.