Home Research
Home Research

Sparsity based regularization and variable selection

Research Area: Learning Theory and Algorithms
Status: In progress  
Faculty: Ernesto De Vito, Alessandro Verri Participants: Lorenzo Rosasco, Silvia Villa, Matteo Santoro


An important goal in many learning problems is to identify the variables relevant to ensure good prediction properties from data points sampled from a very high dimensional space. Under the assumption that the number of relevant variables is usually quite small, we investigate


  • the design of suitable regularization terms promoting sparse solutions in optimization problems
  • the development of efficient optimization algorithms
  • the study of the prediction properties, including consistency, of the proposed solution