October 2020
Spotlight Summary by Xavier Porte
Impact of optical coherence on the performance of large-scale spatiotemporal photonic reservoir computing systems
Optics offers unique advantages to scale-up the number of connections in artificial neural networks, significantly reducing the computational loads related to operate large coupling matrices. However, its implementation rises a fundamental question: does optical coupling come with a price for the system’s performance? In this article, R. M. Nguimdo et al. address this important question by comparing the performance of similar coherent and incoherent spatiotemporal photonic neural networks in several benchmark tasks for image and video recognition. The coupling scheme between reservoir neurons being optical in the coherent case and digital (implemented by an auxiliary computer) in the incoherent case. Crucially, the authors find that the incoherent configuration exhibits a larger memory capacity and higher image and video processing capacities than the coherent configuration, leading to lower classification error rates. Such results call attention to optimize the performance of novel hardware schemes for artificial neural networks, searching a balance between claimed advantages and final performance.
You must log in to add comments.
Add Comment
You must log in to add comments.
Article Information
Impact of optical coherence on the performance of large-scale spatiotemporal photonic reservoir computing systems
Romain Modeste Nguimdo, Piotr Antonik, Nicolas Marsal, and Damien Rontani
Opt. Express 28(19) 27989-28005 (2020) View: Abstract | HTML | PDF