Entanglement assisted training algorithm for supervised quantum classifiers

Abstract

We propose a new training algorithm for supervised quantum classifiers. Here, we have harnessed the property of quantum entanglement to build a model that can simultaneously manipulate multiple training samples along with their labels. Subsequently, a Bell-inequality-based cost function is constructed, that can encode errors from multiple samples, simultaneously, in a way that is not possible by any classical means. We show that upon minimizing this cost function one can achieve successful classification in benchmark datasets. The results presented in this paper are for binary classification problems. Nevertheless, the analysis can be extended to multi-class classification problems as well.