Benchmarking superconducting qubits with generative model learning

Our work is focused on the identification and development of simple machine learning tasks that can act as hardware benchmarks to compare the relative performance of NISQ devices. Using MMD training and stochastic optimization of circuit parameters, we show how a recently introduced class of generative models (the Quantum Circuit Born Machine [1]) can quantify the performance of noisy superconducting qubits. We identify three sources of error that limit the performance of these models on noisy qubits: decoherence, gate fidelities and measurement errors. We construct several shallow depth circuit ansatz and using metrics which are related to fidelity we demonstrate how different errors affect model performance. We also investigate the effect of applying error mitigation to the final trained circuit versus incorporating error mitigation into the circuit training workflow.

[1] Liu, Jin-Guo, and Lei Wang. "Differentiable learning of quantum circuit Born machine." arXiv:1804.04168 (2018).