Optimisation of Quantum Machine Learning Circuit for Classification, Using Variational Quantum Classifier

Abstract

Quantum machine learning (QML) is an emerging area of research in the domain of Quantum computing, leveraging unique properties of entanglement and superposition. Although QML is in early stages of development, yet there are quite many areas where its potential advantage is highly probable, ranging from finance to drug discovery and molecular simulations. Quantum computers of NISQ(Noisy Intermediate Scale Quantum Computers) era are prone to decoherence and errors. Its is expected that the novelty of the quantum computing in general and QML in particular will be more evident with fault tolerant quantum computers in future. Variational Quantum Eigen-solver (VQE) is an algorithm in QML that replicates the classical neural network employing both quantum and classical resources. We took supervised machine learning approach employing the instance of VQE called the Vari aitonal Quantum Classifier(VQC); a classification algorithm. Feature map and Ansatz are important components of VQC for mapping classical data into quantum states and serving as a parameterised circuit to look for ground state for the Hamiltonian of the system, respectively. Number of the repetitions of both featuremap and ansatz applied in a circuit significantly fluctuates the accuracy of the model. Following thesis look for the optimal number of repetitions. For the coding implementation a library ’qiskit’ is used in the python framework. Two data sets ’Penguins data’ and ’Titanic data’ are used with ZZfeature map for feature mapping and Real-mplitude as an ansatz. Number of repetitions for feature map has been varied from 1 to 3 and for ansatz it has been varied from 1 through 5. Various combinations help determine the optimal number of repetitions and our findings establish that less number of repetitions of featuremap and more number of repetitions of ansatz give better accuracy.