Total: 1
Among all the physical platforms for the realization of a Quantum Processing Unit (QPU), neutral atom devices are emerging as one of the main players. Their scalability, long coherence times, and the absence of manufacturing errors make them a viable candidate.. Here, we use a binary classifier model whose training is reformulated as a Quadratic Unconstrained Binary Optimization (QUBO) problem and implemented on a neutral atom QPU. In particular, we test it on a Credit Card Fraud (CCF) dataset. We propose several versions of the model, including exploiting the model in ensemble learning schemes. We show that one of our proposed versions seems to achieve higher performance and lower errors, validating our claims by comparing the most popular Machine Learning (ML) models with QUBO SVM models trained with ideal, noisy simulations and even via a real QPU. In addition, the data obtained via real QPU extend up to 24 atoms, confirming the model's noise robustness. We also show, by means of numerical simulations, how a certain amount of noise leads surprisingly to enhanced results. Our results represent a further step towards new quantum ML algorithms running on neutral atom QPUs for cybersecurity applications.