Optical neural networks with new non-linear activation functions for phase recovery applied to digital holography

• Grant number: 22/15276-0
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: March 01, 2023
• End date: October 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Physics - General Physics
• Principal Investigator: Niklaus Ursus Wetter
• Grantee: Felipe Maia Prado
• Host Institution: Instituto de Pesquisas Energéticas e Nucleares (IPEN). Secretaria de Desenvolvimento Econômico (São Paulo - Estado). São Paulo , SP, Brazil
• Associated scholarship(s): 25/05245-9 - Optomechanical Resonators in Delay Systems for Photonic Reservoir Computing, BE.EP.DR

Abstract

Artificial neural networks are capable of solving complex problems, such as facial recognition, object detection and phase recovery. However, the computational cost is higher when compared to other artificial intelligence algorithms, making high-performance computers necessary. One solution is to develop neural networks using photonic computing, as it is unparalleled in its ability to interconnect and communicate in terms of bandwidth, which can negate the bandwidth and interconnectivity trade-offs inherent in electronic systems. Light communications are known to significantly reduce latency and signal attenuation, as well as increase multiplexing possibilities and transmission capacity due to their propagation speed. In less than four years since the creation of the first all-optical neural network, the scarcity of optical activation functions makes it difficult to advance in the production of faster and more efficient neural networks. Here we propose the creation of optical neural networks using physical non-linear activation functions based on optical devices of wide knowledge of our group, such as passive Q-switching, second harmonic generation, Raman emission, and random lasers inside and outside the Anderson location.

Scientific publications

(5)

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

PRADO, FELIPE MAIA; MIHO DE SOUZA, PEDRO HENRIQUE; DA SILVA, SIDNEY LEAL; WETTER, NIKLAUS URSUS; IEEE. Photoelastic Dispersion Coefficient by Holographic Reconstruction with Neural Networks and the Fresnel Method. 2023 INTERNATIONAL CONFERENCE ON OPTICAL MEMS AND NANOPHOTONICS, OMN AND SBFOTON INTERNATIONAL OPTICS AND PHOTONICS CONFERENCE, SBFOTON IOPC, v. N/A, p. 2-pg., 2023-01-01. (22/15276-0)

PRADO, FELIPE MAIA; FRANCO, TOMAS JUNQUEIRA; WETTER, NIKLAUS URSUS; IEEE. Highly-efficient, dual-wavelength Nd:YLF laser emitting at 1314 nm and 1047 nm. 2023 INTERNATIONAL CONFERENCE ON OPTICAL MEMS AND NANOPHOTONICS, OMN AND SBFOTON INTERNATIONAL OPTICS AND PHOTONICS CONFERENCE, SBFOTON IOPC, v. N/A, p. 2-pg., 2023-01-01. (22/15276-0, 17/10765-5)

DA SILVA, SIDNEY L.; PRADO, FELIPE M.; BRITO, ISIS, V; SOGA, DIOGO; GOMES, LIGIA F.; WETTER, NIKLAUS U.; MURAMATSU, MIKIYA. Holographic method for stress distribution analysis in photoelastic materials. AIMS MATERIALS SCIENCE, v. 11, n. 4, p. 14-pg., 2024-01-01. (12/18162-4, 19/23700-4, 22/15276-0)

PRADO, FELIPE MAIA; FRANCO, TOMAS JUNQUEIRA; WETTER, NIKLAUS URSUS. Efficient Trichromatic Nd:YLF Laser Emitting at 1047 nm, 1053 nm and 1314 nm. PHOTONICS, v. 10, n. 10, p. 10-pg., 2023-10-01. (22/15276-0, 12/11437-8)

PRADO, FELIPE MAIA; FRANCO, TOMAS JUNQUEIRA; WETTER, NIKLAUS URSUS. 24 mJ Single Pulse Energy from a Nd: YLF/Cr:YAG Laser in a DBMC Configuration. 2024 SBFOTON INTERNATIONAL OPTICS AND PHOTONICS CONFERENCE, SBFOTON IOPC 2024, v. N/A, p. 3-pg., 2024-01-01. (17/10765-5, 22/15276-0)