Interactive machine learning from multiple biodata sources

Precision medicine, which is tailored to individual patients, brings a difficult challenge to data analysis methods: The goal is to learn to predict effectiveness of treatments using data with extremely small, even equal to one patient, sample size but an extremely large number of variables. Additional data exist but their relevance is unknown, access may be restricted due to privacy, and the complex models required for good predictivity are computationally heavy. Prof. Kaski develops computational, so-called machine learning methods for analysing multiple data sources, taking into account tacit knowledge of domain experts with interactive modelling. Along with the increasing data-intensivity, similar problems have become very general in other fields of science and services, starting from information retrieval.