Date |
April 21, 2011 |
Speaker |
Dr. Motoaki Kawanabe, Fraunhofer FIRST and TU Berlin, Germany. |
Title |
Classifying Visual Objects with many Kernels |
Abstract |
Combining information from various image features has become
a standard technique in object recognition tasks. However, the
optimal way of combining the resulting kernel functions is usually
unknown in practical applications. Multiple kernel learning (MKL)
techniques allow to determine an optimal linear combination of such
similarity matrices. Classical approaches to MKL promote sparse
mixtures. Unfortunately, these so-called 1-norm MKL variants are often
observed to be outperformed by an unweighted-sum kernel. The
contribution of this paper is twofold: Firstly, we apply a recently
developed non-sparse MKL variant to state-of-the-art object
recognition tasks. We study whether non-sparsity helps in situations
where uniform and sparse mixtures are prone to fail. Secondly,
starting from kernel target alignment, we develop a strategy to reduce
the number of kernels in scenarios where learning using all available
kernels is infeasible. We report on empirical results using the PASCAL
VOC 2008 data.
|
|