Informative Component Extraction with Robustness Consideration

Small sample size of training data might bring trouble as the bias of the estimated parameters for a pattern recognition system. Plug-in test statistics suffer from large estimation errors, often causing the performance to degrade as the measurement vector dimension increases. The informative component extraction method helps to solve this problem by throwing out some dimensions which have relative small distance to the nominal model in statistic sense. Preserving the discriminative information for identification increases the performance. Considering the distortion of the estimated distribution, we introduce the idea of robustness in the informative component extraction. A tolerance ball is applied in the selection of informative and robust components for each individual model (hypothesis). When dealing with multiple parameters model, the supreme of all tolerance balls is used. Informative component extraction with robustness consideration could be used in nonparametric density case simply with slight modification. We use two methods to extract informative component and the performance is examined with 4 different training data sets. The simulation results are compared and discussed with improved performance when considering the robustness.