The current state of the art in the fabrication of quantum processors corresponds to the so-called Noise Intermediate-Scale Quantum (NISQ) era, characterized by devices operating with tens to a few hundred qubits. Such devices are too small and not advanced enough (yet) to leverage the expected theoretical quantum primacy and guarantee fault-tolerance. Nevertheless, a new class of algorithms may provide the first instances of quantum advantage. The latter, called NISQ algorithms, rely on developing small-sized parametric/variational quantum circuits, and find applications in several scientific disciplines, like chemistry or Quantum-enhanced Machine Learning (QML). The present seminar discusses the application of a supervised QML technique, called Quantum Support Vector Machine (QSVM) (a quantum-enhanced version of SVM) to separate signal and background events in the challenging analysis of the 0νββ decay of the Xe-136 isotope in Xenon-doped Lar TPC experiments. The TPC measurements preprocessing involved both traditional (Convolutional Neural Network) and more innovative (Attention Network) deep learning models, comparing both methods in terms of standalone classification power as well as feature extractors for QSVM. Quantum Support Vector Machines have been exhaustively explored for the task, first by designing ad hoc quantum circuits, and later by carrying out a meta-heuristic search with a genetical generative algorithm. The final results highlight how the QSVM algorithms can be successfully run on an IBM quantum processor and reach the same signal/background discrimination performances as the best SVM classifiers.
Seminario CFP, hibrido: Edificio 2, Sala María de Maeztu / Zoom
Coordenadas zoom: https://cern.zoom.us/j/98768015328, pass 092020