Basis Selection from Multiple Libraries

• Activity Number: 505
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Yuedong Wang*+
• Companies: University of California, Santa Barbara
• Keywords: model selection

Abstract

We propose a new method for estimating complex functions by linear combinations of basis functions selected adaptively from different classes of basis functions called libraries. Libraries are chosen to model various features of a function such as change points and oscillations. Data-driven estimates of model complexities based on the generalized degrees of freedom are used to correct bias incurred by adaptive model selection. The proposed method is general in the sense that it can be applied to any generic libraries including spline and wavelet bases. Simulations and real data sets will be used for illustration.