Book/Master Thesis

FZJ-2023-00822

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Denoising with Quantum Machine Learning

Pazem, J. (Corresponding author)FZJ*

2022
Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag Jülich
ISBN: 978-3-95806-641-0

Please use a persistent id in citations: http://hdl.handle.net/2128/33794  urn:nbn:de:0001-2023013111

Abstract: This master thesis explores aspects of quantum machine learning in the light of an application to dampen the effects of noise on NISQ processors. We investigate the possibility of designing machine learning models that can be accommodated entirely on quantum devices without the help of classical computers. With Dissipative Quantum Neural Networks, we simulate a quantum feed-forward neural network for denoising: the Quantum Autoencoder. We assign it to correct bit-flip noise in states that can exist only quantum mechanically, namely the highly entangled GHZ-states. The numerical simulations report that the QAE can recover the target states up to some tolerance threshold on the noise intensity. To understand the limitations, we investigate the mechanisms behind the completion of the denoising task with quantum entropy measures. The observationsreveal that the latent representation is key to reconstructing the desired state in the outputs. Consequently, we propose an inexpensive modification of the original QAE: the brain box-enhanced QAE. The addition of complexity in the intermediate layers of the network maximizes the robustness of the QAE in a setting where only a finite-size training data set is available. We close the argument with a discussion on the generalization properties of the network.

Note: Masterarbeit, RWTH Aachen University, 2022

---

Contributing Institute(s):

1. Theoretische Nanoelektronik (PGI-2)
2. Theoretische Nanoelektronik (IAS-3)

Research Program(s):

1. 5224 - Quantum Networking (POF4-522) (POF4-522)

Appears in the scientific report 2022

---

Database coverage:
;

---