Quantum optical neural networks (QONNs) are reconfigurable nonlinear photonic circuits that take inspiration from neural networks to perform quantum information processing. Implemented on a mature photonic platform, QONNs behave as near-perfect quantum optical devices. Yet, previous models of QONNs have been based on idealized circuits that do not suffer from losses or weak optical nonlinearities. Here, we use the example of a Bell-state analyzer, a device that can detect or generate quantum entanglement, to describe the limitations of realistic QONNs. We explain how these networks operate, then show how their fidelity depends on photon loss and weak nonlinearities, including how these imperfections can be balanced via the choice of network size. We show how this balance provides a way to design near-optimal, brain-inspired quantum photonic solutions to key roadblocks of viable quantum technologies.

The collaborators of this work include J. Carolan, B. J. Shastri, and N. Rotenberg. This work was supported by Natural Sciences and Engineering Research Council of Canada, the Canadian Foundation for Innovation, the Vector Institute and Queen's University.