Résumé : In this talk, I will review several machine learning tasks that fall under the general umbrella of predicting compatibility score F(x,y) of a pair of objects x and y, where both objects are allowed to come from arbitrary sets. Predicting structured output, link prediction in networks as well as multi-variate association analysis in paired datasets falls under this umbrella. I will demonstrate how kernel methods can be used to tackle such modelling problems on medium-scale datasets which are frequent in life science applications. 

I will review applications that involve small molecules as one of the object types of interest: Metabolite identification (Brouard et al. 2017), drug bioactivity prediction (Cichonska et al. 2018) and genotype-phenotype association analysis (Cichonska et al. 2016).
Bio: Juho Rousu is a Professor of Computer Science at Aalto University, Finland. Rousu obtained his PhD in 2001 from University of Helsinki, while working at VTT Technical Centre of Finland. In 2003-2005 he was a Marie Curie Fellow at Royal Holloway University of London. In 2005-2011 he held Lecturer and Professor positions at University of Helsinki, before moving to Aalto University in 2012. Rousu’s main research interest is in developing principled machine learning methods, models and tools for computational and data science. Rousu’s favourite topics include learning with multiple and structured targets, multiple views and ensembles, with methodological emphasis in regularised learning, kernels and sparsity, as well as efficient convex/non-convex optimisation methods. His applications of interest include metabolomics, biomedicine, pharmacology and synthetic biology.
Calendrier des séminaires