Number of projects:9
The analysis of tree rings leads to multiple information on trees and on their environment. In particular, one can determine the age of the tree, the climatic conditions during the growth of the tree, the mechanical stresses that were exerted on the tree as well as the impact of natural or human induced stresses. This project aims at testing the micro-CT method to identify and analyze the distributional patterns of rings for different tree species in relation to climatic changes. This goal of this project is to design and to implement an image-analysis pilot with the major aims to: 1) design a methodology to analyze the tree rings from micro-CT images 2) test this methodology on the tree rings of selected tree specimens and 3) report on the pros and cons of this methodology in comparison with present practices. This project is interdisciplinary and will be supervised by a team composed of scientists from the UNIGE and EPFL: Charlotte Grossiord (EPFL ENAC IIE PERL), Markus Stoffel (UNIGE, DESTE), Daniel SAGE (EPFL STI IMT LIB) and Pascal Turberg (EPFL ENAC IIC PIXE).
In generative adversarial networks, it has been shown that controlling the Lipschitz regularity of a network largely improves the generative performance. For example, Wasserstein GAN (WGAN) and Spectral Normalization GAN (SNGAN) achieve this by restricting the discriminative function to be 1-Lipschitz. Recently, we developed a framework for learning activations of deep neural networks with the motivation of controlling the global Lipschitz constant of the input-output relation. The goal of this project is to investigate the effect of our framework in the generative adversarial networks within various setups. The student should have solid programming skills, in particular being familiar with PyTorch and a general understanding of the main concepts of deep learning.
References Aziznejad, S., Gupta, H., Campos, J., & Unser, M. (2020). Deep neural networks with trainable activations and controlled Lipschitz constant. arXiv preprint arXiv:2001.06263.
Particle fields include a large range of samples of interest, such as bubbles, droplets, or biological cells.
To obtain a three-dimensional (3D) volume of such fields, one popular method involves in-line digital holography (DH).
In this imaging modality, the particle field is illuminated with an incident field (light) so that multiple scattering and diffraction occur. The resulting field is then holographically recorded.
From a single two-dimensional (2D) DH image, computational methods are able to recover the particles within a 3D volume. When the density of particles and/or the depth of field are large, the reconstruction task becomes too difficult for conventional methods.
During this project, the student will implement and train a neural network to recover particles within a 3D volume from a 2D image. The programming language is Python (Pytorch). Based on an existing code in Matlab, the student will also implement the physical model which describes the wave propagation in Pytorch. The required skills are prior knowledge of deep learning, proficiency in coding in Pytorch. The student should be able to learn the basics of wave propagation and optics during the project.
During this project, the student will understand the physical model of an imaging modality, learn how to conduct a complete project with deep learning, and learn how to use a physical model combined with deep learning.
References Tahir, W., Kamilov, U. S., & Tian, L. (2019). Holographic particle localization under multiple scattering. Advanced Photonics, 1(3), 036003.
In computed tomography (CT), the goal is to reconstruct a 3D object from a set of its 2D projections. Typically, this reconstruction task is formulated as an optimization problem where one exploits certain properties of the signal of interest (e.g., sparsity in a transform domain). However, over the past decade, several learning-based methods have been shown to outperform the classical reconstruction methods. In this project, we consider a setting where relatively fewer projections are available and the idea is to use generative adversarial networks (GANs) for the reconstruction task. The student should be familiar with the PyTorch framework and should have a general understanding about basic deep learning concepts. Prior experience in inverse problems and/or optimization is a definite plus.
References:  H. Gupta, K.H. Jin, H.Q. Nguyen, M.T. McCann, M. Unser, "CNN-Based Projected Gradient Descent for Consistent CT Image Reconstruction"  A. Bora, A. Jalal, E. Price, A. G. Dimakis, "Compressed sensing using generative models"
Single-particle cryo-electron microscopy (cryo-EM) has revolutionised the field of structural biology over the last decade, culminating in 2017 by the awarding of the Nobel Prize in Chemistry to its three founders. Nowadays, single-particle cryo-EM permits the regular discovery of new biological structures at atomic resolution. Yet, the reconstruction task remains an enduring challenge due to the unknown orientations adopted by the 3D particles prior to imaging. The goal of this project is to further strengthen a recently-developed joint optimization scheme that efficiently alternates between the reconstruction and the estimation of the unknown orientations . More precisely, the student will introduce a multiscale scheme  inside the iterative-refinement framework itself to benefit from the robustness gained by reconstructing volumes at coarser scales. The student should have a strong interest in image processing, and good Matlab skills are a prerequisite. An interest in inverse problems and/or optimization theory is a definite plus. References:  M. Zehni, L. Donati, E. Soubies, Z. Zhao, M. Unser, "Joint Angular Refinement and Reconstruction for Single-Particle Cryo-EM," IEEE Transactions on Image Processing, vol. 29, pp. 6151-6163, 2020.  L. Donati, M. Nilchian, C.Ó.S. Sorzano, M. Unser, "Fast Multiscale Reconstruction for Cryo-EM," Journal of Structural Biology, vol. 204, no. 3, pp. 543-554, December 2018.
The simplex algorithm is the oldest but still one of the most popular optimization algorithms for solving linear programs. In short, it iterates through vertices of the feasible region of the linear program until it reaches a minimum of the cost function. Recently, it was been shown that it can be used to reach an extreme point of the solution set of inverse problems with l1 regularization . This is relevant because these extreme point solutions are guaranteed to be sparse, i.e., they can be expressed with few nonzero coefficients . Different methods can be applied to solve inverse problems with l1 regularization using the simplex, with different tradeoffs in terms of problem dimension, speed and perhaps numerical stability. The goal of this project is to benchmark these different methods in order to quantify the aforementioned tradeoffs. The student should have a strong interest in optimization.
 Gupta, Harshit, Julien Fageot, and Michael Unser. "Continuous-Domain Solutions of Linear Inverse Problems with Tikhonov versus Generalized TV Regularization." IEEE Transactions on Signal Processing 66.17 (2018): 4670-4684.
 Unser, Michael, Julien Fageot, and Harshit Gupta. “Representer Theorems for Sparsity-Promoting l1 Regularization”, IEEE Transactions on Information Theory 62.9 (2016): 5167-5180.
High speed imaging can be achieved with any camera, as has recently been demonstrated with the Virtual Frame Technique. Using this method, any monotonic phenomenon that is imaged instantaneously with perfect contrast (such that the image is binary) can be recorded at high rates by increasing the exposure time of the camera. Thus, complex temporal dynamics can be recorded at high rates and high resolution, averting the traditional trade-off between size of the region of interest and imaging rate. In this project we explore the limits of this method, using a dual-approach of experimental demonstration and theoretical analysis. Project in collaboration with Engineering Mechanics of Soft Interfaces (EMSI) laboratory (https://www.epfl.ch/labs/emsi/).
Classification and clustering are some of the most important objectives in supervised and unsupervised learning, respectively. Interestingly, in both scenarios, the learning scheme eventually produces a piecewise-constant function. This remarkable property allows one to analyze them jointly. The goal of this project is to develop a variational framework to estimate piecewise-constant functions and to derive an efficient learning algorithm, built as a module. One can then also use this module in deep neural networks and compare the performance with classical setups for various applications of classification and clustering.
Blebbing is a very dynamic phenomenon that plays an important role during apoptosis, cell migration, or cell division. Using time-lapsed microscopy techniques, phase contrast and fluorescence, biologists can observe blebs which are spherical protrusions which appear and disappear on the membrane of the cell. The goal of the project is to design and to implement image-analysis algorithms based on active contour and curve optimization take into account the blebbing. It requires a automatic segmentation of the cell over the multichannel sequence of images and a local extraction of the bulges to quantify blebbing. The project will be implemented in Java as an ImageJ plugin with an user interface allowing a manual edition of the outlines of the blebs.