A Novel Solution Scheme for the Kernel MSE Model
Jinghua Wang · 2009
In this paper we first show that the minimum squared-error solutions of kernel minimum squared-error (KMSE) models are neither unique nor numerically stable. We then propose a novel scheme for KMSE. This solution scheme can produce the unique solution and the maximum between-class margin. This solution has a highly accurate classification rate. The experimental result illustrates the feasibility and effectiveness of MNMSE solution of KMSE.