$l_p$-Norm Multiple Kernel Learning

With Prof. Ulf Brefeld, Technical University of Dortmund

Abstract:

Learning linear combinations of multiple kernels is an appealing strategy when the right choice of features is unknown. Previous approaches to multiple kernel learning (MKL) promote sparse kernel combinations to support interpretability and scalability. Unfortunately, this $l_1$-norm MKL is rarely observed to outperform trivial baselines in practical applications. To allow for robust kernel mixtures that generalize well, we extend MKL to arbitrary norms. We devise new insights on the connection between several existing MKL formulations and develop two efficient interleaved optimization strategies for arbitrary norms, that is $l_p$-norms with p>=1. This interleaved optimization is much faster than the commonly used wrapper approaches, as demonstrated on several data sets. A theoretical analysis and an experiment on controlled artificial data shed light on the appropriateness of sparse, non-sparse and $l_{\infty}$ norm MKL in various scenarios. Importantly, empirical applications of $l_p$-norm MKL to three real-world problems from computational biology show that non-sparse MKL achieves accuracies that surpass the state-of-the-art.

Short Bio:

Ulf Brefeld recently joined the Technical University of Dortmund. Prior to that he worked at University of Bonn, Yahoo! Research Barcelona and in machine learning groups at Technische Universität Berlin, Max Planck Institute for Computer Science in Saarbrücken and Humboldt-Universität zu Berlin. He received a Diploma in Computer Science in 2003 from Technische Universität Berlin and a Ph.D. (Dr. rer. nat.) in 2008 from Humboldt-Universität zu Berlin. He works in statistical machine learning and data mining. This includes learning from structured data, kernel methods, semi-supervised techniques, information extraction/retrieval, and applications in natural language processing and computational biology.

Date: Monday October 22th, 2012, 10:00am

Location: Battelle bât D, room D185

5 octobre 2012

À la Une