Garcia Saez, ArturHuanay de Dios, Álvaro Ari2023-02-222023-02-222022-07https://hdl.handle.net/2445/193932Màster Oficial de Ciència i Tecnologia Quàntiques / Quantum Science and Technology, Facultat de Física, Universitat de Barcelona. Curs: 2021-2022. Tutor: Artur Garcia SaezAdversarial attacks on discriminative algorithms are highly used in the field of cybersecurity (e.g. on email-filtering or malware bypassing). As the automationof tasks is since now used more than ever, all current state-of the art attack and defence methods are frequently exploited by some sort machine learning action. This project aims to build a pipeline to bypass a discriminator (ResNet-18) using a Probabilistic Generative Model (Restricted Boltzmann Machine) by generating images as seemingly as possible to real hand-written numbers (MNIST dataset) trained by two different methods; classical machine learning and quantum-enhanced machine learning (classical machine learning with a final boost of quantum annealing) using D-Wave´s quantum computers. Quantum computing is proposed as an alternative to minimize more the free energy of the model at the end of the training. Computation of the loss function is proven to be easier with a quantum annealing machine rather than with fully classical methods. A better accuracy is expected by the quantum-enhanced model as well as a faster training. Quality of the images generated by each technique is compared and possible applications in the field of cybersecurity using PGMs are proposed besides of discussing physical requirements24 p.application/pdfengcc-by-nc-nd (c) Huanay, 2022http://creativecommons.org/licenses/by-nc-nd/3.0/es/Aprenentatge automàticSeguretat informàticaOrdinadors quànticsTreballs de fi de màsterMachine learningComputer securityQuantum computersMaster's thesesinfo:eu-repo/semantics/masterThesisinfo:eu-repo/semantics/openAccess