Regularization parameter estimation for spectral regression discriminant analysis based on perturbation theory

Spectral regression discriminant analysis (SRDA) is an important subspace learning method. It has a tunable parameter, i.e., the regularization parameter, which critically affects the performance. However, how to set this parameter automatically has not been well solved to date. In SRDA, this regularization parameter was only set as a constant, which is usually suboptimal. In this paper, we develop a new algorithm to automatically estimate the regularization parameter of SRDA based on the perturbation linear discriminant analysis (PLDA). Experiments on multiple data sets demonstrate the effectiveness of the proposed method.