Section: New Results

Structured Sparsity Regularization & Statistical Hypothesis Testing

Paticipants: Eugene Belilovsky, Wacha Bounliphone, Katerina Gkirtzou, Andreas Argyriou, Matthew Blaschko

We have developed novel methods for structured sparsity regularization & hypothesis testing. We have applied these methods to fMRI [3] , [2] , [36] and the analysis of large medical databases [10] . We have also developed novel statistical hypothesis tests for relative dependency [21] , [37] and similarity [14] . We have applied these methods to the problem of identifying relative dependencies between languages using a multi-lingual corpus, and for discovering the relative relationships between gliomas and genetic information. Additionally, we have shown the application of relative tests to the problem of model selection in deep generative models, and currently an important question in machine learning.