Polariton neural networks | Prof. Michał Matuszewski скачать в хорошем качестве

Polariton neural networks | Prof. Michał Matuszewski

Timecodes are below the abstract. Prof. Michał Matuszewski Institute of Physics, Polish Academy of Sciences Title: “Efficient optical computing with exciton-polaritons” Abstract Recent years have witnessed remarkable developments in big data, artificial intelligence and neural networks. Machine learning has found wide applications in both research and the industry. This comes at the cost of high levels of energy consumption that are necessary to process large amounts of data. It is expected that over 20% of global electricity use by 2030 will be used for information processing. The performance of complementary metal-oxide semiconductors (CMOS) no longer follows Moore's law. In result, much research has been aimed at finding an alternative platform for information processing, characterized by high performance and energy efficiency. In this talk I will review recent progress in machine learning with photons. Photonic information processing benefits from high speed, parallelization, low communication losses, and high bandwidth. Fully functional photonic neurons, including spiking neurons, as well as neural networks, have been already realized in laboratories. Several networks achieved high performance in challenging machine learning tasks, such as image and video recognition. We recently demonstrated hardware neural network systems where strong optical nonlinearity results solely from interactions of exciton-polaritons, quantum superpositions of light and matter [1,2,3]. Such superpositions, in the form of mixed quasiparticles of photons and excitons, are characterized by excellent photon-mediated transport properties and strong exciton-mediated interactions. These semiconductor microcavity systems can be used to construct fully all-optical neural networks characterized by extremely high energy efficiency [4]. We show why using polaritonics in place of standard nonlinear optical phenomena, is the key to achieving such a performance. Main paper/arXiv, related to the seminar, and other references [1] A. Opala, S. Ghosh, T. C. Liew, and M. Matuszewski, Physical Review Applied 11 , 064029 (2019) [2] D. Ballarini, A. Gianfrate, R. Panico, A. Opala, S. Ghosh, L. Dominici, V. Ardizzone, M. De Giorgi, G. Lerario, G. Gigli, Timothy C. H. Liew, Michal Matuszewski, and Daniele Sanvitto, Nano Letters 20, 3506 (2020) [3] R. Mirek, A. Opala, P. Comaron, M. Furman, M. Król, K. Tyszka, B. Seredynski, D. Ballarini, D. Sanvitto, Timothy C. H. Liew, Wojciech Pacuski, Jan Suffczyński, Jacek Szczytko, Michał Matuszewski, and Barbara Piętka, Nano Letters (2021) [4] M. Matuszewski, A. Opala, R. Mirek, M. Furman, M. Król, K. Tyszka, T.C.H. Liew, D. Ballarini, D. Sanvitto, J. Szczytko, B. Piętka, arXiv:2108.12648. 0:00:00 Intro 0:00:25 Start of the talk 0:01:35 Brief introduction to neural networks 0:08:35 Neuromorphic computing 0:11:05 Electrons and photons 0:12:17 Early experiments on optical neural networks 0:16:44 Optical neural networks 0:19:21 Optical reservoir computing 0:20:35 Polariton neural networks 0:21:29 Exciton-polaritons 0:24:16 Reservoir computing 0:26:18 Reservoir computing with exciton-polaritons 0:34:15 Experiments: reservoir computing 0:36:17 Experiments: binarized neural network 0:41:11 Polariton neural networks:outlook 0:45:47 Conclusions and outlook 0:46:23 Questions The School of Physics and Engineering of ITMO University (Saint Petersburg, Russia) hosts online weekly seminars to discuss recent scientific achievements of researchers from all over the world with a wide audience. The topics include experimental (Optical seminar) and theoretical (Theoretical seminar) aspects of nanophotonics, solid-state physics, material science, as well as design and applications of radiofrequency structures and devices (Microwave seminar). Upcoming and past events can be found at https://physics.itmo.ru/en/seminars   / physics.itmo     / physics_itmo     / physics.itmo   #TheoreticalSeminar #ITMO #новыйфизтех #scientificseminar