Actor neural networks for the robust control of partially measured nonlinear systems showcased for image propagation through diffuse media

The output of physical systems, such as the scrambled pattern formed by shining the spot of a laser pointer through fog, is often easily accessible by direct measurements. However, selection of the input of such a system to obtain a desired output is difficult, because it is an ill-posed problem; that is, there are multiple inputs yielding the same output. Information transmission through scattering media is an example of this problem. Machine learning approaches for imaging have been implemented very successfully in photonics to recover the original input phase and amplitude objects of the scattering system from the distorted intensity diffraction pattern outputs. However, controlling the output of such a system, without having examples of inputs that can produce outputs in the class of the output objects the user wants to produce, is a challenging problem. Here, we propose an online learning approach for the projection of arbitrary shapes through a multimode fibre when a sample of intensity-only measurements is taken at the output. This projection system is nonlinear, because the intensity, not the complex amplitude, is detected. We show an image projection fidelity as high as similar to 90%, which is on par with the gold-standard methods that characterize the system fully by phase and amplitude measurements. The generality and simplicity of the proposed approach could potentially provide a new way of target-oriented control in real-world applications when only partial measurements are available.

Learning of physical systems: from inference to control

Babak Rahmani

To characterize a physical system to behave as desired, either its underlying governing rulesmust be known a priori or the system itself be accurately measured. The complexity of fullmeasurements of the system scales with its size. When exposed to real-world conditions, suchas perturbations or time-varying settings, the system calibrated for a fixed working conditionmight require non-trivial re-calibration, a process that could be prohibitively expensive, inefficientand impractical for real-world use cases.In this thesis, a learning procedure for solving highly ill-posed problems of modeling a system'sforward and backward response functions is proposed. In particular, deep neural networksare used to infer the input of a system from partial measurements of its outputs or to obtain adesired target output from a physical system.I showcase the applicability of the proposed methods for inference and control in opticalmultimode fibers. Amplitude/phase-encoded input of a multimode fiber is reconstructedfrom intensity-only measurements of the outputs. Conversely, the required input of the fiberfor projecting a desired output is obtained using intensity-only measurements of the output.Next, the stochastic neural network of the retina in Salamander is modeled by a probabilisticneural network. The model is used to optimize the input stimuli so as to find the simplestspatiotemporal patterns that elicit the same neuronal spike responses as those elicited byhigh-dimensional stimuli.As demonstrated in this thesis, application of data-driven methods for characterization ofcomplex large-scale real-world systems has proved useful in simplifying the measurement apparatus,end-to-end optimization of the system and automatic compensation of perturbation.

EPFL2022

Imaging through multimode fibers using deep learning: The effects of intensity versus holographic recording of the speckle pattern

Navid Borhani, Eirini Kakkava, Georgia Konstantinou, Damien Claude-Marie Loterie, Christophe Moser, Demetri Psaltis, Babak Rahmani, Ugur Tegin

Information transmission through multimode fibers (MMFs) has been a topic of great interest for many years. Deep learning algorithms have been applied successfully to MMFs in particular to fiber endoscopy. In this work, we show how Deep Neural Networks (DNNs) can be a versatile technique for classification and recovery of input images that have been significantly distorted while propagating along the MMF forming a speckle pattern. A comparison between holographic and intensity-only recording of the speckle output, which is used as an input to the DNNs, shows that high performance can be achieved without having the full field information (amplitude and phase). Impressive reconstruction fidelity and classification accuracy of the fiber inputs from the intensity-only images of the speckle patterns is reported.

ELSEVIER SCIENCE INC2019

Linear Recursive State Estimation of Hybrid and Unbalanced AC/DC Micro-Grids using Synchronized Measurements

Willem Lambrichts, Mario Paolone

In this paper, we present an exact (i.e. non-approximated) and linear measurement model for hybrid AC/DC micro-grids for recursive state estimation (SE). More specifically, an exact linear model of a voltage source converter (VSC) is proposed. It relies on the complex VSC modulation index to relate the quantities at the converters DC side to the phasors at the AC side. The VSC model is derived from a transformer-like representation and accounts for the VSC conduction and switching losses. In the case of three-phase unbalanced grids, the measurement model is extended using the symmetrical component decomposition where each sequence individually affects the DC quantities. Synchronized measurements are provided by phasor measurement units and DC measurement units in the DC system. To make the SE more resilient to vive step changes in the grid states, an adaptive Kalman Filter that uses an approximation of the prediction-error covariance estimation method is proposed. This approximation reduces the computational speed significantly with only a limited reduction in the SE performance. The hybrid SE is validated in an EMTP-RV time-domain simulation of the CIGRE AC benchmark micro-grid that is connected to a DC grid using 4 VSCs. Bad data detection and identification using the largest normalised residual is assessed with respect to such a system. Furthermore, the proposed method is compared with a non-linear weighted least squares SE in terms of accuracy and computational time.

2022