Prof. Christopher Moser

Christophe Moser is associate professor of Optics in the Microengineering department at EPFL. He obtained his PhD at the California Institute of Technology in optical information processing in 2000. He co-founded and was the CEO of Ondax Inc (acquired by Coherent Inc.), Monrovia California for 10 years before joining EPFL in 2010. His current interests are ultra-compact endoscopic optical imaging through multimode fibers, retinal imaging, additive manufacturing via volumetric 3D printing with light. He cofounded Composyt light lab in the field of head worn displays in 2014 (acquired by Intel Corp). He is the author and co-author of 70 peer reviewed publications and 40 patents.

Website – https://lapd.epfl.ch/ 

Talk

Matrix vs Machine Learning for transmitting images through multimode fibers

Abstract

Direct image transmission in multimode fibers (MMFs) is hampered by modal scrambling inside the fiber due to the multimodal nature of the medium. To undo modal scrambling, approaches that either use interferometry to construct a transmission matrix or iterative adaptive optics to form an output spot on the camera have been proposed and implemented successfully. The Matrix method is a calibration that consists of measuring experimentally the phase and amplitude of the light at the output of the fiber, by interferometry, for a number of orthogonal patterns approximately equal to the number of modes of the fiber. By inverting the matrix, the complex field at the input is computed that gives the desired intensity pattern at the output of the fiber. In this way, the MMF behaves as a projector.
For a practical implementation, an intensity detection rather than interferometric detection is preferred during the calibration phase. The tradeoff for the simpler detection scheme is to solve a non-linear inverse problem. I will show recent results on using Artificial neural networks to approximate the non-linear mapping between the input and output of the fiber and discuss the pros and cons of machine learning versus a matrix approach.