Boosting the Validity of Multi-Class Quantum Outputs: Living on the Edge

TL;DR

This paper proposes a new measurement mapping for multi-class quantum classifiers that improves output validity and accuracy with fewer samples, addressing quantum noise and statistical limitations in quantum machine learning.

Contribution

It introduces a novel qubit measurement mapping to the edges of an n-dimensional simplex, enhancing output quality and reducing sample requirements in VQCs.

Findings

Improved output validity over one-hot encoding.

Enhanced classification accuracy with fewer samples.

Better robustness against quantum noise.

Abstract

Quantum machine learning (QML) aims to use quantum computers to enhance machine learning, but it is often limited by the required number of samples due to quantum noise and statistical limits on expectation value estimates. While efforts are made to reduce quantum noise, less attention is given to boosting the quality of the discrete outputs from Variational Quantum Classifiers (VQCs) to reduce the number of samples needed to make confident predictions. This paper focuses on output representations in multi-class classification, introducing a new mapping of qubit measurements to edges of an n-dimensional simplex, representing independent binary decisions between each pair of classes. We describe this mapping and demonstrate how it offers a direct improvement to the number of valid circuit output samples as well as the accuracy of those outputs over one-hot encoding while advocating for few-sample accuracy as a primary goal for effective VQCs.