BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Pablo Bermejo DTSTART:20240924T140000Z DTEND:20240924T150000Z DTSTAMP:20260423T021418Z UID:TalentQ/10 DESCRIPTION:Title: Quantum Convolutional Neural Networks are (Effectively) Classically Simul able\nby Pablo Bermejo as part of Quantum Spain\n\n\nAbstract\nQuantum Convolutional Neural Networks (QCNNs) are widely regarded as a promising model for Quantum Machine Learning (QML). In this work we tie their heuris tic success to two facts. First\, that when randomly initialized\, they ca n only operate on the information encoded in low-bodyness measurements of their input states. And second\, that they are commonly benchmarked on "lo cally-easy'' datasets whose states are precisely classifiable by the infor mation encoded in these low-bodyness observables subspace. We further show that the QCNN's action on this subspace can be efficiently classically si mulated by a classical algorithm equipped with Pauli shadows on the datase t. Indeed\, we present a shadow-based simulation of QCNNs on up-to 1024 qu bits for phases of matter classification. Our results can then be understo od as highlighting a deeper symptom of QML: Models could only be showing h euristic success because they are benchmarked on simple problems\, for whi ch their action can be classically simulated. This insight points to the f act that non-trivial datasets are a truly necessary ingredient for moving forward with QML. To finish\, we discuss how our results can be extrapolat ed to classically simulate other architectures.\n LOCATION:https://researchseminars.org/talk/TalentQ/10/ END:VEVENT END:VCALENDAR