Quantum Generative Adversarial Networks

Introduction

In our last years of high school, in Catalonia, we are required to do a research project. In mine, I explored the world of Quantum Machine Learning. By the end of the project, I had programmed a functional Quantum Generative Adversarial Network for image generation, and written a very extensive document (~120 pages) explaining all my work from scratch. This included the basics in both Quantum Machine Learning and traditional Neural Networks. I loved working on this project, I am really passionate about the research field as well.

Abstract

Quantum Machine Learning is one of the most promising applications of Quantum Computing. In this work, I present a Quantum Generative Adversarial Network (qGAN) for generating gray-scale bar images. Through the Fréchet Distance score, I evaluate the effectiveness of a partial measurement on the simulated quantum circuit that generates the images. This score shows that the measurement improves qGAN performance by avoiding an oscillation on the resemblance between the generated and the real images, that is, an alternation of good and poor quality generated images that occurs through optimization of the qGAN. For queries to the author open an issue.

Project Report Document

It is written in catalan, the oficial language of Catalonia, students are required to write in it.