Enhancing Qubit Readout with Autoencoders

TL;DR

This paper introduces a neural network-based qubit readout method using autoencoders to improve classification accuracy in superconducting qubit measurements, especially for challenging short and long time signals.

Contribution

It presents a novel autoencoder pre-training approach for neural networks to enhance superconducting qubit readout classification performance.

Findings

Improved classification accuracy for short measurement times.

Enhanced performance for long measurement signals.

Outperforms traditional readout methods in key scenarios.

Abstract

In addition to the need for stable and precisely controllable qubits, quantum computers take advantage of good readout schemes. Superconducting qubit states can be inferred from the readout signal transmitted through a dispersively coupled resonator. This work proposes a novel readout classification method for superconducting qubits based on a neural network pre-trained with an autoencoder approach. A neural network is pre-trained with qubit readout signals as autoencoders in order to extract relevant features from the data set. Afterwards, the pre-trained network inner layer values are used to perform a classification of the inputs in a supervised manner. We demonstrate that this method can enhance classification performance, particularly for short and long time measurements where more traditional methods present lower performance.