Hybrid Quantum Machine Learning Classifier with Classical Neural Network Transfer Learning

Abstract

The hybrid model uses a minimally trained classical neural network as a pseudo-dimensional pool to reduce the number of features from a data set before using the output for forward and back propagation. This allows the quantum half of the model to train and classify data using a combination of the parameter shift rule and a new “unlearning rate” function. The quantum circuits were run using Penny-Lane simulators. The hybrid model was tested on the wine data set to promising results of up to 97% accuracy. The inclusion of quantum computing in machine learning represents great potential for advancing this area of scientific research because quantum computing offers the potential for vastly greater processing capability and speed. Quantum computing is currently primitive; however, this research takes advantage of the mathematical simulators for its processing that prepares this work to be used on actual quantum computers as soon as they become widely available for machine learning. Our research discusses a benchmark of a Classic Neural Network as made hybrid with the Quantum Machine Learning Classifier. The Quantum Machine Learning Classifier includes the topics of the circuit design principles, the gradient parameter-shift rule and the unlearning rate. The last section is the results obtained, illustrated with visual graphs, with subsections on expectations, weights and metrics.