Learning Quantum Error Models

In this abstract we propose a methodology for learning quantum error models from experimental data. This information is useful for characterizing the effectiveness of hardware, predicting how well a circuit should run in practice, and synthesizing corrected circuits that attempt to perform better by taking the error model into account.

We learn the error model by taking each gate in the original circuit and replacing it with a parameterized probability distribution over potential gates. For example, we could replace a Pauli X gate with a distribution having probability p of performing a random unitary and probability 1-p of performing the Pauli X. We then perform bayesian inference to deduce the most likely error model that gave us the desired error.

We test our methodology on experimental data, and evaluate the learned error models in its predictive power.