FisMat2017 - Submission - View

Abstract's title: Experimental benchmark of Boson Sampling with pattern recognition techniques
Submitting author: Iris Agresti
Affiliation: La Sapienza, Università di Roma
Affiliation Address: Piazzale Aldo Moro 5, I-00185 Roma
Country: Italy
Oral presentation/Poster (Author's request): Oral presentation
Other authors and affiliations: Niko Viggianiello (La Sapienza, Università di Roma, Italy), Fulvio Flamini (La Sapienza, Università di Roma, Italy), Nicolò Spagnolo (La Sapienza, Università di Roma, Italy), Andrea Crespi (Istituto di Fotonica e Nanotecnologie, Consiglio Nazionale delle Ricerche (IFN-CNR), Italy), Roberto Osellame (Istituto di Fotonica e Nanotecnologie, Consiglio Nazionale delle Ricerche (IFN-CNR), Italy), Nathan Wiebe (Station Q Quantum Architectures and Computation Group, Microsoft Research, United States), Fabio Sciarrino (La Sapienza, Università di Roma, Italy)

The difficulty of validating large-scale quantum devices, such as Boson Samplers, poses a major challenge

for any research program that seeks to show quantum advantages over classical hardware. Boson Sampling is a

computational problem formally defined by Aaronson and Arkhipov in 2011 that has been shown to be classically

intractable (even approximately) under mild complexity theoretic assumptions. Thus, demonstrating that a

quantum device can efficiently perform Boson Sampling is powerful evidence that quantum computing can bring

exponential advantages over its classical counterpart. However, despite the fact that Boson Sampling is within our

reach, its measurement statistics are intrinsically exponentially hard to predict, so that the validation of a Boson

Sampler is not a straightforward task for large quantum systems.

To address this problem, we propose a novel data-driven approach wherein models are trained to identify common

pathologies using unsupervised machine learning algorithms. The aim is to find an inner structure in an unknown

data set through clustering, grouping data in different classes according to collective properties recognized by the

algorithm. We illustrate this idea by training a classifier that uses K-means clustering to distinguish between

Boson Samplers that use indistinguishable photons from those that do not. We train the model on numerical simulations

of small-scale Boson Samplers and then validate the pattern recognition technique on larger numerical

simulations as well as on integrated photonic platforms in both traditional Boson Sampling experiments and scattershot

experiments. This approach performs substantially better on the test data than previous methods and

underscores the ability to further generalize its operation beyond the scope of the examples it was trained on.