Choisir la langue :

J. Salmon (Univ. Montpellier): Safe screening rules to speed-up sparse regression solvers

Institutional tag: 
Thematic tag(s): 

In high dimensional regression scenarios, sparsity enforcing penalties have proved useful to regularize the data-fitting term. A recently introduced technique called screening rules, leverage the expected sparsity of the solutions by ignoring some variables in the optimization, hence leading to solver speed-ups. When the procedure is guaranteed not to discard features wrongly the rules are said to be "safe". The proposed framework can cope with generalized linear models and various sparsity enforcing penalty functions, though the talk will focus mainly on least-squares with l1 regularization for simplicity.
Our Gap Safe rules (so called because they rely on duality gap computation) allow to safely discard more variables than previously considered safe rules, particularly for low regularization parameters. We can handle any iterative solver but our contribution is particularly well suited for (block) coordinate descent techniques. We report significant speed-ups compared to previously proposed safe rules both on simulated and real data.

Dates: 
Friday, May 24, 2019 - 11:00 to 12:00
Location: 
Inria B11
Speaker(s): 
Joseph Salmon
Affiliation(s): 
Universiyé de Montpellier