Imaging Science Community Meets in Bologna

The SIAM Activity Group on Imaging Science held the 9th SIAM Conference on Imaging Science (IS18) in Bologna, Italy from June 5 to 8, 2018. The University of Bologna, located in the beautiful historical center and regarded as the oldest university in the world, hosted the conference and served as the perfect backdrop for a stimulating meeting. IS18 was attended by a record number of over 800 participants (almost double that of 2016) from across the world,representing academia, industry, and government laboratories. The conference chairs, Fiorella Sgallari (University of Bologna) and Omar Ghattas (University of Texas at Austin), were particularly pleased to welcome over 171 graduate students. With the help of an outstanding team of students and staff, the local organizer, Fiorella Sgallari, kept this immense meeting well on track and exceeded all expectations by organizing multiple social activities, including city tours, several lunch and evening receptions, a banquet featuring traditional Bolognese cuisine and live music, and the social dinner in the magical atmosphere of Palazzo Re Enzo, located in the heart of Medieval Bologna.

The scientific program of the four-day meeting was packed and spanned new advances in mathematical and computational techniques in imaging and its diverse applications, as well as a focus on recent advances in machine learning. The five plenary speakers, whose slides are available, also reflected this diversity of the field of imaging science. Raymond Chan (Chinese University of Hong Kong) opened the event and presented new robust powerful methods for segmenting noisy color images, for hyper-spectral image classification and point cloud segmentation. Anna Michalak's (Stanford University) presentation identified mathematical challenges in imaging atmospheric greenhouse gas emissions and outlined potential impacts on climate policy. Christoph Schnoerr (University of Heidelberg) gave an overview of past and current major trends in image segmentation and presented recent advances made by new models based on partial differential equations. Yonina Eldar's (Technion) talk showed how mathematical signal processing techniques allow high-resolution imaging with minimal measurements and overviewed improvements made in a wide range of applications including but not limited to medical imaging. Addressing the imaging communities’ pressing needs for efficient techniques for optimization and processing large datasets, Francis Bach (Ecole Normale Superieure) gave an overview of recent advances in stochastic optimization methods. 

In addition, there were three mini-tutorials on the topics of computational methods for uncertainty quantification (John Bardsley, University of Montana), regularization of inverse problems (Otmar Scherzer, University of Vienna), and 3D reconstruction from satellite images (Gabriele Facciolo, École Normale Supérieure Paris-Saclay). Slides for the minitutorials can be accessed here. In a forward-looking plenary discussion on the future of imaging in the age of machine learning, panelists Gitta Kutyniok (TU Berlin), Peyman Milanfar (Google Research), Rosemary Renaut (Arizona State University), and Joachim Weickert (Saarland University) identified the merging of model-based and data-based methodologies as one key focus for future research. They also called for the need to develop rigorous theoretical underpinnings for machine learning methods such as deep neural networks. A further point of discussion was the training of experts in machine learning methods for imaging science. 

The plenary and tutorial sessions were complemented by 153 minisymposia, 65 posters, and 11 contributed sessions. Organization of this wide-ranging program was made possible by the contributions of the scientific committee consisting of 12 leading researchers in imaging science and the 20 members of the organizing committee from six Italian universities. The lecture and poster contributions spanned a wide range of topics, and although it is impossible to give a representative overview in this article, a short summary of a few representative minisymposia selected by the organizers and SIAG-IS board is given here. 

A two-part minisymposium organized by Arvind Saibaba (North Carolina State University), Julianne Chung, and Eric de Sturler (both of Virginia Tech) featured recent developments on Krylov methods for large-scale inverse problems, data assimilation, and uncertainty quantification. The sessions highlighted the critical role of iterative techniques for solving inverse problems that arise in many important imaging applications such as image deblurring and tomographic reconstruction. Showing that image data is more than pixels and voxels, the minisymposium organized by Martin Storath (Universität Heidelberg), Martin Holler (École Polytechnique, Université Paris Saclay), and Andreas Weinmann (Hochschule Darmstadt) considered general geometric data and new variational approaches. The speakers presented various applications in science and engineering in which data are not given in a vector space but in a nonlinear space such as a manifold. Interesting examples are circle and sphere-valued data as appearing in synthetic-aperture radar imaging and the space of positive matrices with the Fisher-Rao metric, which is the underlying data space for Diffusion Tensor Imaging. 

The minisymposium entitled “Imaging with Light and Sound,” organized by Felix Lucka (CWI Amsterdam) and Tanja Tarvainen (University of Eastern Finland)  brought together leading researchers in the fields of optical, acoustic, and coupled imaging. The talks identified increasing interest in imaging with photons in the visible and near-infrared spectrum, which due to the physiological nature of chromophores in tissue allows one to obtain a rich set of contrasts. However, this commonly comes at the cost of complex models of light propagation and usually poor resolution resulting from the strongly ill-posed and nonlinear nature of the corresponding inverse problem. The speakers presented  remedies for these barriers  by combining optical and other modalities, e.g., light-plus-sound modalities. 

The sessions organized by Yaniv Romano (Technion - Israel Institute of Technology), Peyman Milanfar (Google Research), and Michael Elad (The Technion - Israel Institute of Technology) examined image denoising from a new angle. Image denoising has been studied for decades, yet it continues to be an active and important field. The speakers showed that its influence lies beyond the denoising application, as new algorithms frequently serve as building blocks for tackling more general low-level computer vision problems, and recently have been shown to serve as effective regularizers for solving a wide range of inverse problems. This minisymposium highlighted the current state of the art in image denoising, and unique roles it can play in solving general inverse problems for applications both inside and outside of imaging science. 

Several mathematical approaches for and new insights into machine learning were presented  in a three-part minisymposium organized by Michael Moeller (University of Siegen) and Gitta Kutyniok (Technische Universität Berlin). Motivated by the recent success of learning-based approaches for imaging problems, the speakers outlined new avenues for analysing and interpreting the involved machine learning models by applying a wide array of analysis techniques from approximation and regularization theory, optimization, and differential equations.

In an awards session on the last day of the conference, the Best Challenging Application Award was given to Claudio Belvedere for his presentation “Multi-instrument Medical Imaging Analysis for Personalized Joint Replacement Design.” “A Sequential Monte Carlo for Astronomic Imaging” by Federica Sciacchitano received the Best Poster Award, with second place going to Maria Oliver for “L1 Patch-Based Image Partitioning Into Homogeneous Textured Regions” and third place to Antoine Houdard for “How to Use Mixture Models on Patches for Solving Image Inverse Problems.” 

In their business meeting, the SIAG-IS laid out a roadmap for the next Imaging Science meeting and beyond. Keeping to the biennial schedule, the next meeting will be held jointly with the annual meetings of SIAM and the Canadian Applied and Industrial Mathematics Society (CAIMS) on July 6-10, 2020 in Toronto, Canada. Motivated by the highly successful overseas meetings in Hong Kong and Bologna, there was broad support for the idea of alternating the conference between locations within North America and outside. Looking forward, this will require an international contingent who is willing to undertake the organization of the 2022 IS conference.