SELECT is an INRIA team involved in statistical modeling. We solve model selection problems in a statistical learning framework, focusing on hidden structure models, statistical pattern recognition, and statistical decision problems in a global sense. Applications include reliability, high-dimensional biological data analysis, and big data in general.
SELECT proposes tools and software for model and variable selection, often based on penalized likelihood criteria. Usually, we take a non-asymptotic view using concentration inequalities, or a Bayesian one that can take into account the modeling context.