Ulugbek Kamilov (MERL, USA)

Computational Image Reconstruction under Multiple Scattering

Abstract: Multiple scattering of an electromagnetic wave as it passes through an object is a fundamental problem that limits the performance of current imaging systems. From the perspective of imaging inverse problems, multiple scattering leads to nonlinear forward models that generally lead to intractable optimization problems. In this talk, I will discuss recent advances for designing optimization schemes that can account for multiple scattering while also accommodating model-based priors for imaging.

Biography: Ulugbek S. Kamilov is a Research Scientist in the Computational Sensing team at Mitsubishi Electric Research Laboratories (MERL), Cambridge, MA, USA. Dr. Kamilov obtained his B.Sc. and M.Sc. in Communication Systems, and Ph.D. in Electrical Engineering from the École Polytechnique Fédérale de Lausanne (EPFL), Switzerland, in 2008, 2011, and 2015, respectively. In 2007, he was an Exchange Student at Carnegie Mellon University (CMU), Pittsburgh, PA, USA, in 2010, a Visiting Student at Massachusetts Institute of Technology (MIT), Cambridge, MA, USA, and in 2013, a Visiting Student Researcher at Stanford University, Stanford, CA, USA. Dr. Kamilov is a member IEEE Special Interest Group on Computational Imaging since January 2016.

Dr. Kamilov's research focus is computational imaging with an emphasis on the development and analysis of large-scale computational techniques for biomedical and industrial applications. His research interests cover imaging through scattering media, multimodal imaging, distributed radar sensing, and through-the-wall imaging. He has co-authored 17 journal and 36 conference publications in these areas. His Ph.D. thesis work on Learning Tomography (LT) was selected as a finalist for EPFL Doctorate Awards 2016 and was featured in the "News and Views" section of the Nature magazine.

References:

• [1] Kamilov, Papadopoulos, Shoreh, Goy, Vonesch, Unser, and Psaltis, “Learning Approach to Optical Tomography,” Optica, 2015.
• [2] Zhang, Godavarthi, Chaumet, Maire, Giovannini, Talneau, Allain, Belkebir, and Sentenac, "Far-field diffraction microscopy at lambda/10 resolution,” Optica, 2016.
• [3] Liu, Liu, Mansour, Boufounos, Waller, and Kamilov, “SEAGLE: Sparsity-Driven Image Reconstruction under Multiple Scattering,” arXiv:1705.04281 [cs.CV], 2017.

Gabriel Peyré (ENS, Paris, France)

Optimal Transport and Deep Generative Models

Abstract: In this talk, I will review some recent advances on deep generative models through the prism of Optimal Transport (OT). OT provides a way to define robust loss functions to perform high dimensional density fitting using generative models. This defines so called Minimum Kantorovitch Estimators (MKE) [1]. This approach is especially useful to recast several unsupervised deep learning methods in a unifying framework. Most notably, as shown respectively in [2,3] (and reviewed in [4]) Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN) can be interpreted as (respectively primal and and dual) approximate MKE. This is a joint work with Aude Genevay and Marco Cuturi.

Biography: Gabriel Peyré graduated from Ecole Normale Supérieure de Cachan, France, in 2003 and received his Ph.D in applied mathematics from Ecole Polytechnique, Paris, France, in 2005. From 2006 to 2016, he has been a researcher at the Centre Nationale de Recherche Scientifique (CNRS), working in Ceremade, University Paris-Dauphine. Since 2016, he is a CNRS Research Director at Ecole Normale Supérieure de Cachan. He was head of the research group SIGMA-Vision, which was funded by the European Research Council (ERC). SIGMA-Vision activities were focussed on sparse and adaptive representations with application in computer vision, computer graphics and neurosciences. Since 2016, he leads his new ERC project called NORIA, on "Numerical Optimal tRansport for ImAging". Since 2005 Gabriel Peyré has co-authored 65 papers in international journals, 8 book chapters, 2 books, 74 conference proceedings in top vision and image processing conferences, and two books. He is the creator of the "Numerical tour of signal processing" (www.numerical-tours.com), a popular online repository of Matlab/Scilab resources to teach modern signal and image processing.

References:

• [1] Federico Bassetti, Antonella Bodini, and Eugenio Regazzini. On min- imum Kantorovich distance estimators. Statistics & probability letters, 76(12):1298–1302, 2006.
• [2] Olivier Bousquet, Sylvain Gelly, Ilya Tolstikhin, Carl-Johann Simon-Gabriel, and Bernhard Schoelkopf. From optimal transport to generative modeling: the VEGAN cookbook. Arxiv:1705.07642, 2017.
• [3] Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein GAN. Arxiv:1701.07875, 2017.
• [4] Aude Genevay, Gabriel Peyré, Marco Cuturi, GAN and VAE from an Optimal Transport Point of View, Arxiv:1706.01807, 2017