Abstract
Many problems in imaging have been significantly improved by the recent developments of deep learning. However, the mathematical analysis of deep learning is still lagging behind its impressive practical performance. In particular, the modeling aspects of deep learning often do not follow a clear rationale, but are mostly based on heuristics coupled with exhaustive experimentation.The aim of this session is to present and discuss the most recent approaches that combine and connect existing well-founded mathematical approaches with deep learning.

More on 'Detailed programme'

Abstract
Solving inverse problems has always called for ways to stabilize the problem, reduce sensitivity to errors in measurements, and to constrain the space of solutions. The standard tools for this task have for decades been explicit regularization using well-worn norms, which are measured on familiar linear operators acting on the space of solutions. This approach has now run its course -- it’s been recently upended by the creative use of other (nonlinear) operators. Among these, denoisers, and their more specialized cousins proximal operators, have played a starring role. These have opened a new door that allows us to work with more freedom to control stability, and to flexibly constrain our computational approaches to solving inverse problems. This is the subject of this session.

More on 'Detailed programme'

Abstract
Modern imaging methods increasingly rely on the Bayesian statistical framework to solve challenging imaging problems. That is, they use stochastic models to represent the data observation process and the prior knowledge available, and they obtain solutions by using inference techniques stemming from Bayesian decision theory, delivering accurate and insightful results. Applying Bayesian strategies to imaging problems is not straightforward, and this drives the development of new methods and algorithms that tightly combine ideas from signal processing, stochastics, computational statistics, optimisation, numerical analysis, and beyond. This session will present a range of exciting new developments in Bayesian analysis and computation methodology for solving imaging problems.

More on 'Detailed programme'

Abstract
Machine learning has gained incredible popularity across many disciplines in recent years owing to its high flexibility and performance on real world problems. However, its use in high precision science can be problematic owing to the “black box” nature of many algorithms and the difficulty in quantifying uncertainties and systematic biases. This session brings speakers from diverse fields within astronomy, with a focus on novel uses of machine learning and its responsible application to obtain high precision scientific constraints.

More on 'Detailed programme'

Abstract
On the path towards the Square Kilometre Array, numerous pathfinder and precursor instruments have been or are being developed. The exciting scientific possibilities of these new telescopes pose a number of imaging challenges, such as imaging of very large fields-of-view, high-dynamic range imaging, online processing and autonomous data quality assessment, that go beyond the capabilities of traditional radio astronomical image formation methods. This session provides an overview of new avenues being explored to address these challenges and allows for in-depth discussion of recent progress and future directions.

More on 'Detailed programme'

Abstract
Large surveys are increasingly dominating the astrophysics landscape. The amount of data provided by these surveys open up many new opportunities for learning about the Universe with statistical methods, but they also cause challenges stemming from noise, observational biases, known and unknown pipeline defects, systematics, etc. This session will focus on the opportunities and challenges of doing astrostatistics in the era of large surveys, with a focus on novel tools for handling and analyzing large-survey data now and in the future.

More on 'Detailed programme'

Abstract
Machine learning and artificial intelligence are expected to play an increasingly important role in our healthcare system, and in particular in imaging. While these technologies are usually associated with developments that aim to extract diagnostic information from medical images, research activities with the goal of using machine learning for image reconstruction have picked up significantly over the last two years. The presentations in this session will cover novel core machine learning developments like model architectures and learning algorithms, as well as application to MRI and CT reconstruction.

More on 'Detailed programme'

Abstract
Biomedical imaging is currently experiencing an exciting era of new methodological developments. In MRI, compressed sensing, numerical physical models and machine learning are promoting faster, quantitative and comprehensive examinations. In CT, the ideas of sparsity and deep learning combined with new collimator and detector hardware are enabling efficient uses of radiation dose and new contrasts. In PET, similar advances in hardware and image reconstruction algorithms are underway. Though many of these methods have been developed independently in each of the different imaging modalities, their combination may be seen as example of a new paradigm of rapid, comprehensive, and information-rich tomography. This session will explore cross-cutting themes in each modality and will attempt to promote transfer of ideas between investigators in different areas of medical imaging.

More on 'Detailed programme'

Abstract
Over the past two decades, there has been a significant paradigm shift in bioimaging research. Increasingly, researchers are coming to consensus that computation is a cornerstone of future imaging systems. The goal of this session is to bring together experts working in different computational imaging modalities including optical tomography, hyperspectral microscopy, optical coherence tomography, and ultrasound.

More on 'Detailed programme'