EF1 – Extracting dynamical Laws from Complex Data



Sparsity and Sample-Size Efficiency in Structured Learning

Project Heads

Sebastian Pokutta

Project Members

David Martínez-Rubio

Project Duration

01.01.2022 − 31.12.2023

Located at



In this project, we will investigate the properties of sparse solutions induced through learning algorithms that promote sparsity. In particular, we want to look at sparsity as being a consequence of model and the implicit sparsity bias of the chosen optimization method. Closely connected to this is the question of sample-size efficiency as these algorithms also tend to have a much higher sample-size efficiency by implicitly restricting the degrees of freedom.

We will also look at algorithms that are able to quickly approximate the optimal fair resource allocation, given a notion of fairness in some problems. For instance, under the fairness axioms of Pareto optimality, independence of irrelevant alternative, affine invariance and symmetry, the optimal fair allocation can be posed as an optimization problem. This direction of research concerns the study of finding allocations that approximate this or other fair allocations efficiently.

External Website

Related Publications

Related Pictures