Abstract
We consider the problem of learning a linear combination of pre-specified kernel matrices in the Fisher discriminant analysis setting. Existing methods for such a task impose an ¿1 norm regularisation on the kernel weights, which produces sparse solution but may lead to loss of information. In this paper, we propose to use ¿2 norm regularisation instead. The resulting learning problem is formulated as a semi-infinite program and can be solved efficiently. Through experiments on both synthetic data and a very challenging object recognition benchmark, the relative advantages of the proposed method and its ¿1 counterpart are demonstrated, and insights are gained as to how the choice of regularisation norm should be made.