Axis 2 – Monitoring and Prediction of Complex Systems:

Optical hardware neural networks chasing for the performance of human brain

Neuromorphic computing, whose principles are inspired by the human brain, is a revolutionary analog paradigm for artificial intelligence, able to solve complex computational tasks with performances that digital computer science struggles or even fails to provide. It has been applied to benchmark problems such as speech or image recognition, or processing and prediction/classification of complex multi-parameter signals, demonstrating excellent potential for applications like financial forecasting, electroencephalography or fuel cell system control. EIPHI researchers are among the pioneers in its hardware development, creating intelligent optoelectronic processors fully following neuromorphic design principles. In such advanced physical implementations, individual neurons (e.g. a pixel of a light modulator or micro-lasers in a laser array) are nonlinearly interconnected into a so-called optical neural network, which after a trained readout learning stage can autonomously process the incoming data. Sheler Maktoobi, an EIPHI doctorate student, has dedicated herself to address the challenge of creating large reservoirs of potentially millions of neuronal photonic interconnects, optimizing their connectivity and learning. Recognizing her excellent work, she was recently awarded the Best Poster Presentation at the 6th International Symposium in Optics and its Applications of the SPIE Student Chapter at Trento, Italy, Feb. 2018, for her contribution “Diffractive coupling for Optical Neural Networks”.