Title: “Squeezing the most out of ridge”

Speaker: Professor Mark van de Wiel, Amsterdam University Medical Center

Abstract: Ridge regression is nowadays regarded as a somewhat old-fashioned technique: most statisticians prefer sparse models, while machine learners use non-linear, multi-layer prediction algorithms. The aim of this talk is to endorse ridge by illustrating its strengths, from a methodological, computational and applied perspective, including:

1. Flexibility (in terms of output and type of covariates)
2. Bayesian counterpart facilitates empirical Bayes estimation of penalty parameter(s)
3. Differential penalization for multi-modal (e.g. multi-omics) data
4. Penalty weights moderated by (external) information on the features:
   1. Co-data
5. Effective posterior feature selection
6. Computational efficiency
7. Performance in clinical (omics) prediction problems

These strengths will be illustrated on a cancer multi-omics prediction problem.