Image reconstruction through unknown random configurations of multimode fibers using deep learning

[S. Resisi et al, arxiv, 2011.05144 (2020)]

Multimode fibers tend to scramble input images due to intermodal dispersion and random mode coupling. While this scrambling effect can be learned, for instance by measuring the transmission matrix of the fiber, slight changes of the geometrical conformation of the fiber modify its response, making the calibration obsolete. In the present paper, the authors use deep-learning using data sets acquired over a wide range of deformations to reconstruct images sent through unknown configurations of multimode fibers. This is possible thanks to the presence of invariant properties that the numerical model learns.

Learning and avoiding disorder in multimode fibers

[M.W. Matthès, arxiv, 2010.14813 (2020)]

In the past 10+ years, numerous advances were made for endoscopic imaging, micromanipulation, or telecommunication applications with multimode fiber. The main limitation to real-life applications is the sensitivity to perturbations that sometimes causes the transmission property of the fiber to change in real-time. To address this issue, the authors (we) show that, even in the presence of strong perturbations, there exists a set of channels that are almost insensitive to perturbations. Interestingly, these channels can be found using only measurements from small perturbation leveraging the so-called generalized Wigner-Smith operator. This requires the measurement of the transmission matrix, which is done thanks to a new technique based on deep learning frameworks that compensate automatically for misalignments and aberrations, allowing fast and easy acquisitions.

Inverse design of planar optical components using deep learning

[N.J. Dinsdale et al., arxiv, 2009.11810, (2020)]

Photonic integrable circuits are basically waveguide structures that allow performing useful operations, such as mode or wavelength multiplexing/demultiplexing in the case of telecommunication applications. For many operations, we can find quite easy solutions, where the shape of the structure imposes certain boundary conditions that force light to behave the way we want. However, for an arbitrary operation, it is not always possible to find a trivial solution. Non-trivial solutions, where the link between the geometry of the structure and its function is not direct, should then be considered. In the present paper, the authors use deep learning to find geometrical configurations for planar photonic circuits that look like disordered waveguides but actually perform a previously chosen linear operation. These configurations lead experimentally to robust, high throughput, and accurate behaviors.

Complex-Valued Neural Networks for Physics Applications

An implementation in PyTorch

Artificial neural networks are mainly used for treating data encoded in real values, such as digitized images or sounds. In such systems, using complex-valued tensors would be quite useless. This is however different for physic related topics. When dealing with wave propagation in particular, using complex values is interesting since the physics typically has linear, hence more simple, behavior when considering complex fields. This is sometimes true even when the inputs and the outputs of the system are real values. For instance, consider a complex media that you excite using an amplitude modulator, such as a DMD (Digital Micromirror Device) and you measure the output intensity. You manipulate only real values, but if you want to characterize the system, you have to keep in mind that the phase is a hidden variable as the effect of propagation is represented by the multiplication by a complex matrix on the optical field.