Abstract
Bilinear inverse problems such as blind deconvolution have been studied by mathematicians and engineers for a long time. Recently, there has been increased interest in such problems, as randomization has been incorporated into the measurement models, allowing for stronger recovery guarantees. Despite a number of promising works, the study of bilinear and mulilinear problems from this new viewpoint, similar to the one in compressed sensing, is only in its beginning. This session will present recent progress and will discuss further research directions.

More on 'Detailed programme'

Abstract
In a wide range of application fields in signal/image processing (satellite imaging, astronomy, medical imaging, seismic data processing, chemistry, biophysics ...), inverse problems need to be solved so as to recover to the data of interest. Their resolution usually requires to solve very high dimensional optimization problems, involving neither necessarily differentiable nor convex functionals. Recently, increasingly powerful tools have been introduced such as proximal methods (e.g. augmented Lagrangian formulations), or primal-dual approaches. But their application to real-world inverse problems remains a challenging task, due to the very large size of the data to be handled. In particular, it may become necessary to take advantage of modern multicore or distributed computing architectures. The objective of this special session is to bring together researchers from the applied mathematics and signal/image processing communities so as to present novel powerful methods paving the way towards a new generation of algorithms for solving inverse problems at a large scale.

More on 'Detailed programme'

Abstract
Modern research in imaging inverse problems (ranging from simple denoising to tomographic reconstruction or magnetic resonance imaging) has been largely dominated by sparsity-based regularization, equivalently, maximum a posteriori (MAP) Bayesian criteria with sparsity-inducing priors. The goal of this session is to bring together a group of researchers who are producing exciting new work outside of this "dominating paradigm", namely by adopting other inference criteria (such as the Bayesian posterior expectation, which achieves the minimum mean squared error) or developing learning-based approaches that do not rely on sparsity-based generative models of the underlying images.

More on 'Detailed programme'

Abstract
The session will explore novel statistical methodology in the field of astrophysics, from large surveys of the cosmos to the properties of galaxies. Speakers will be covering problems in determining the properties of galaxies, the expansion history, and the large scale distribution of matter in the Universe.

More on 'Detailed programme'

Abstract
Next decade's cosmological data sets from LSST, Euclid, WFIRST and various CMB experiments will pose an exciting opportunity to study the expansion history and structure formation of our Universe. The precision of these data sets requires the development and implementation of new statistical methods to 1) measure the observables in an unbiased way, 2) to account for observational and astrophysical systematics, and 3) to robustly transition from data to model space.

More on 'Detailed programme'

Abstract
The new generation of radio telescopes, building towards the SKA, has reinvigorated the radio interferometry research field over the past decade. Not only has the front-end technology advanced, but the digital nature of these innovative observatories has opened up the field of signal processing for radio interferometry to new techniques and approaches. This session will present recent progress, discuss further research directions and look at emerging issues in the field of next generation radio interferometry.

More on 'Detailed programme'

Abstract
Modern neuroscience has an incredibly rich toolkit at its disposal, with a broad range of techniques that operate at massively different spatial and temporal scales. These techniques are crucial given that brain structure relates to function over at least eight orders of magnitude. A major challenge for neuroscience over the coming decade is to relate these measurements to each other in order to understand how microscopic features relate to whole-brain phenomena. For example, the exploding field of connectomics includes research on defining precise inter-connections between local groups of neurons forming micro-circuits, as well as the long-range pattern of connections between large-scale brain regions. Relating these scales to each other is a major challenge that will require unifying models and detailed experimental work. Similarly, with the advent of population imaging, challenges exist in leveraging increasingly large-n investigations to provide insight at the level of individuals. This session will explore these challenges from the perspective of signal processing, biophysical modelling and data analytics.

More on 'Detailed programme'

Abstract
In recent years, many new techniques for biomedical imaging have been proposed that combine computation and data acquisition in novel ways. Based on the formulation of image reconstruction as an inverse problem, these methods make use of realistic signal models, capture all relevant degrees of freedom in high-dimensional parameter spaces, and exploit prior knowledge about the structure of biomedical images. While there is no shortage of new ideas and methods, practical adoption is laggingbehind. In fact, many new methods are difficult to reproduce without source code, or the published implementation is not robust, not flexible enough, or simply too slow. Significant additional effort is often required to develop a useful implementation. For this reason, the development of generalized reconstruction and acquisition methods and of efficient and versatile software packages plays an increasingly important role in biomedical imaging.

More on 'Detailed programme'

Abstract
In recent times, biomedical imaging has experienced an avalanche of methodological developments that are combining to create new ways of seeing. In the area of optical imaging, previous limits on spatial resolution, volumetric coverage, and dynamic information content have been shattered, resulting in a profusion of biological applications. In MRI, advanced nonlinear and model-based reconstruction algorithms have likewise circumvented old limits, and have begun to change the way imagers acquire and interact with image information. In CT, new reconstructions combined with new classes of detectors are shifting standards for X-ray-based contrast and radiation dose. Similar advances are underway in PET, ultrasound etc. Though many of these developments have arisen independently among the different imaging modalities, taken together they may be seen as example of a new paradigm of rapid, comprehensive, multidimensional, and information-rich tomography. Through a series of modality-specific examples spanning multiple spatiotemporal scales, this session will explore cross-cutting themes, and will attempt to promote transfer of ideas between investigators in disparate areas of imaging.

More on 'Detailed programme'