Quantum generative adversarial networks for data augmentation
Date
2025-12-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Benemérita Universidad Autónoma de Puebla
Abstract
"In this thesis, we address a high-energy physics problem involving two types of events, signal and background, with 50 samples for each where signal events are gluons and background events quarks. Each sample has five features that define the type of event. A quantum generative adversarial network model is proposed, which is trained separately on each type of data. The goal is to replicate the features of each sample, thereby expanding the dataset where sample scarcity was a problem. For this purpose, we propose two very similar codes; the difference between them is that in one, only the samples classified as background are loaded, while in the other, only those classified as signal are loaded, the rest of the architecture is the same. The model is expected to successfully generate samples that imitate the features. If the model is capable of such a thing, then this means it is indeed possible to train a quantum generative adversarial networks model with a small amount of samples to generate realistic data. Another thing that is expected to be proven or refuted with this thesis is that, for this task, a pure quantum model (quantum generator and discriminator) is better than a hybrid one (quantum generator and classic discriminator). With this approach, we aim to demonstrate the usefulness of a quantum architecture in problems where data augmentation is required due to the limited amount of available data".
Description
Keywords
Citation
Collections
Document Viewer
Select a file to preview:
Can't see the file? Try reloading