Quantum error reduction with deep neural network applied at the post-processing stage

TL;DR

This paper introduces a deep neural network-based post-processing method to reduce errors in quantum computations on NISQ devices, leveraging circuit structure without classical simulation data, demonstrated on IBM quantum processors.

Contribution

The proposed approach trains DNNs to correct quantum hardware noise by using artificially increased Trotter steps, without relying on classical simulation data during training.

Findings

Significant error reduction achieved on real quantum hardware.

Method effectively increases quantum circuit depth in Trotter steps.

Applicable to digital quantum simulations of spin chains.

Abstract

Deep neural networks (DNN) can be applied at the post-processing stage for the improvement of the results of quantum computations on noisy intermediate-scale quantum (NISQ) processors. Here, we propose a method based on this idea, which is most suitable for digital quantum simulation characterized by the periodic structure of quantum circuits consisting of Trotter steps. A key ingredient of our approach is that it does not require any data from a classical simulator at the training stage. The network is trained to transform data obtained from quantum hardware with artificially increased Trotter steps number (noise level) towards the data obtained without such an increase. The additional Trotter steps are fictitious, i.e., they contain negligibly small rotations and, in the absence of hardware imperfections, reduce essentially to the identity gates. This preserves, at the training stage, information about relevant quantum circuit features. Two particular examples are considered that are the dynamics of the transverse-field Ising chain and XY spin chain, which were implemented on two real five-qubit IBM Q processors. A significant error reduction is demonstrated as a result of the DNN application that allows us to effectively increase quantum circuit depth in terms of Trotter steps.