Multi-Nyström Method Based on Multiple Kernel Learning for Large Scale Imbalanced Classification

Algorithm 1

The proposed MKLMO algorithm.

	Input. Dataset ; number of landmark points ; rank ; number of kernels ; predefined kernel function .
	Output. Classification result for instance
(1)	Draw subsets of balanced landmark points with each
(2)	Calculate subkernel matrices between the instances in and and among the instances in with kernel
(3)	Calculate the singular value decomposition on
(4)	Approximate according to (8)
(5)	Initialize mixture weights
(6)	While stopping criterion not met do
(7)	Calculate by SVM solver with according to (15)
(8)	Calculate descent direction according to (19)–(21)
(9)	Set
(10)	Whiledo
(11)	Set ,
(12)	Set ,
(13)	Set , ,
(14)	Calculate with
(15)	end while
(16)	Line search along for optimal
(17)	Assign
(18)	end while