Probabilistic Machine Learning and Medical Image Processing (Former Research Group)

The research group "Probabilistic Machine Learning and Medical Image Processing," headed by Dr. Matthias Seeger, was part of the Cluster of Excellence MMCI from October 2008 until December 2010.

Vision and Research Strategy

Acquiring and processing images is at the heart of many medical diagnosis and scientific experimental modalities. In order to cope with explosive growth in demand, tools for acquisition optimization and experimental planning have to make low-level decisions autonomously, based on uncertain knowledge and incomplete data. Bayesian graphical modelling and inference allows for optimal decisions in principle, yet poses hard computational challenges in practice: probability distributions over images have to be processed, capturing degrees of uncertainty about and dependencies within the signal.

We approach these problems by variational calculus, designing convex approximate inference relaxations and algorithmic reductions to scalable estimation technology for the first time. Our framework creates a bridge between notoriously difficult Bayesian decision-making and core computational problems of image reconstruction and numerical mathematics, for which scalable algorithms are known.

Our work ranges from theoretical analysis of variational inference and design optimization over adaptation to novel applications to algorithm design and system development. For the latter, we employ a hybrid structure of graphical model message passing, image reconstruction algorithms and eigensolvers, which will be implemented on GPUs and multi-core hardware.

Composition of Group

The Probabilistic Machine Learning and Medical Image Processing group operated starting in October 2008 within the Cluster of Excellence.

A major focus is on the design, analysis, and implementation of large scale approximate Bayesian inference and signal reconstruction methodology, with applications to image reconstruction and acquisition optimization (compressive sensing) for magnetic resonance imaging and computational photography.

Research Topics and Achievements

Projects and Collaborations

Dr. Matthias Seeger

Dr. Matthias Seeger

From October 2008 until December 2010, Matthias Seeger was heading the Independent Research Group Probabilistic Machine Learning and Medical Image Processing

Fon: +41 21 6931396