Scalable Bayesian Methods for Machine Learning and Imaging

Objective

Machine learning seeks to automatize the processing of
large complex datasets by adaptive computing, a core strategy to meet growing
demands of science and applications.
Typically, real-world problems are mapped to penalized estimation tasks (e.g.
binary classification), which are solved by simple efficient algorithms. While
successful so far, I believe this approach is too limited to
realise the potential of adaptive computing. Most of the work, such as data
selection, feature construction, model calibration and comparison, still has to
be done by hand. Demands for automated decision-making (e.g. tuning
data acquisition during an experiment) are not met.

Such problems are naturally addressed by Bayesian reasoning about uncertain
knowledge, which however remains infeasible in most large scale settings.
The main goal of this proposal is to unite the strengths of penalized
estimation and Bayesian decision-making, exploiting the former's advanced state
of the art in order to implement substantial improvements coming with
the latter in large scale applications. A major focus is on improving magnetic
resonance imaging (MRI) by way of new Bayesian technology, driving robust
nonlinear
reconstruction from less data, and optimizing the acquisition through
Bayesian experimental design, applications not previously attempted by machine
learning. Far beyond the reach of present methodology, these goals demand
a novel computational foundation for approximate Bayesian inference through
numerical algorithmic reductions.

This project will have high impact on probabilistic machine learning, raising
the bar for scalable Bayesian computations. It will help to open up a whole new
range of medical imaging applications for machine learning. Moreover,
substantial impact on MRI reconstruction research is anticipated. There is
strong recent interest in savings through compressive sensing, whose full
potential is realised only by way of adaptive technology such as projected
here.

Fields of science

Funding Scheme

ERC-SG - ERC Starting Grant

Coordinator

ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNE

Address

Batiment ce 3316 station 1
1015 Lausanne
Switzerland

Region

Schweiz/Suisse/Svizzera Région lémanique Vaud

Activity type

Higher or Secondary Education Establishments

Administrative Contact

Caroline Vandevyver (Ms.)

Principal investigator

Matthias Seeger (Dr.)

Links

Contact the organisation Website

EU contribution

No data

Beneficiaries (1)

ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNE

Switzerland