Error mitigation of entangled states using brainbox quantum autoencoders

Abstract

Current quantum hardware is susceptible to various sources of noise that limit our access to precisely-tuned entangled states. Quantum autoencoder circuits with a single-qubit bottleneck have demonstrated the ability to correct errors up to a certain noise threshold. In this work, we introduce subgraph structures in the bottleneck, referred to as brainboxes, and observe not only a speed-up in the denoising process but also the ability to handle stronger noise channels beyond the initial threshold. However, selecting the most suitable brainbox for the bottleneck involves a trade-off between the intensity of noise on the hardware and the training complexity. By monitoring the evolution of Rényi entropy throughout the networks, we present a clear framework for denoising through learning.