Constrained Optimization Methods for Models in Data Fusion 


Workshop, March 27, 2023
Simula Oslo, Room KA23-2-292-Østmarka

In Data Fusion, one analyzes different measurements of the same phenomenon, which are stored in several matrices and/or tensors. The goal is to simultaneously factorize these datasets in order to allow for dimensionality reduction, clustering, feature extraction, prediction and the like. The factorizations are often subject to various constraints which model the type of data at hand. An inherent problem of these models is that they are notoriously hard to solve, especially if several constraints come in conjunction. The aim of this workshop is to learn about existing models and their applications, to study different optimization techniques, and to possibly improve them by introducing more solvers or accelerating the existing ones. One particular focus lies on manifold optimization, where many open problems remain. 

Programme

09:00 - 09:15    Welcome

09:15 - 10:15    Evrim Acar: An introduction to coupled matrix and tensor factorizations

10:15 - 11:00    Discussion and coffee break

11:00 - 12:00    Ronny Bergmann: An introduction to optimization on manifolds

12:00 - 13:30    Lunch Break at Simula Canteen

13:30 - 14:00    Carla Schenker: AO-ADMM framework for regularized coupled matrix-tensor factorizations with linear couplings
14:00 - 14:30    Martin Ludvigsen: Adversarial non-negative matrix factorization for source separation problems

14:30 - 15:00    Discussion and coffee break

15:00 - 15:30    Max Pfeffer: Manifold optimization for data fusion
15:30 - 16:00    Florian Becker: PARAFAC2 for joint analysis of metabolomics data sets

16:00 - 17:30    Discussion and closing

17:30    Pizza at KA23

Registration

You can register for this workshop by filling out this form.

List of Participants