Developing penalized and/or approximate but fast versions of the classic multivariate techniques to produce sparse solutions (involving only a few original variables) applicable for large data sets.
Nickolay Trendafilov leads the research on sparse multivariate methods in the school.