Real-time error mitigation for variational optimization on quantum hardware

In this work, we put forward the inclusion of error mitigation routines in the process of training variational quantum circuit (VQC) models. In detail, we define a real-time quantum error mitigation (RTQEM) algorithm to assist in fitting functions on quantum chips with VQCs. While state-of-the-art QEM methods cannot address the exponential loss concentration induced by noise in current devices, we demonstrate that our RTQEM routine can enhance VQCs’ trainability by reducing the corruption of the loss function. We tested the algorithm by simulating and deploying the fit of a monodimensional 𝑢-quark parton distribution function on a superconducting single-qubit device, and we further analyzed the scalability of the proposed technique by simulating a multidimensional fit with up to eight qubits.