Quantum Machine Learning

Quantum phenomena such as entanglement, tunnelling or superposition are increasingly used in quantum information to perform computation. Moreover, machine learning methods are driving a paradigm shift in the field of artificial intelligence. This project aims to explore the possibility of implementing quantum phenomena into machine learning algorithms. A theoretical approach to achieve this is proposed by modifying the architecture of a classical image recognition neural network such that it imitates the behaviour of a single Q-bit in a magnetic field. The resulting system is a quantum-inspired neural network and does a reasonable job in classifying handwritten digits of 0 and 1.

General Architecture of the proposed Quantum-Inspired-Neural Network

a) The structure of the NN

Note that the input is either a 28x28 pixel handwritten image of 1 or 0. The neurons in the hidden layer 1 correspond to the magnetisation direction of the virtual Q-bit. The neurons in the hidden layer 2 represent the measurement operators.

b) Quantum equivalent of the network

Firstly, the virtual Q-bit is being prepared and subsequently measured. The result of the measurement corresponds to either 1 or 0. Note that the entire system runs on a classical computer but introduces physical/quantum- mechanical phenomena into the ML algorithm.

Referenec: The code is based on the work done by Michael Nielsen, accessible at: https://github.com/mnielsen/neural-networks-and-deep-learning

Update: Further improvements and modifications have been added in collaboration with Ksenija Kovalenka (November 2022).