Sparse Group Lasso: Consistency and Climate Applications

• Activity Number: 24
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Soumyadeep Chatterjee*+ and Karsten Steinhaeuser and Arindam Banerjee and Snigdhansu Chatterjee and Auroop R Ganguly
• Companies: University of Minnesota-Twin Cities and University of Minnesota and University of Minnesota and University of Minnesota and Northeastern University
• Keywords: Sparse Group Lasso ; climate prediction ; statistical consistency

Abstract

The plethora of available climate data gives rise to some unique challenges for designing statistical predictive models, due to its high dimensionality and spatio-temporal nature. The complexity of the data dictates that models should exhibit parsimony in variable selection. Recently, a class of methods which promote structured sparsity in the model have been developed, which is suitable for this task. In our work, we have proved theoretical statistical consistency of m-estimators with hierarchical tree-structured norm regularizers. We consider one particular model, the Sparse Group Lasso (SGL), to construct predictors of land climate using ocean climate variables, in order to capture interesting relationships that exist between land and ocean atmospheric systems. Our experimental results demonstrate that the SGL model provides better predictive performance than the current state-of-the-art, remains climatologically interpretable, and is robust in its variable selection.