Decimated input ensembles for improved generalization

| Using an ensemble of classi ers instead of a single classi er has been demonstrated to improve generalization performance in many di cult problems. However, for this improvement to take place it is necessary to make the classi ers in an ensemble more complementary. In this paper, we highlight the need to reduce the correlation among the component classi ers and investigate one method for correlation reduction: input decimation. We elaborate on input decimation, a method that uses the discriminating features of the inputs to decouple classi ers. By presenting di erent parts of the feature set to each individual classi er, input decimation generates a diverse pool of classi ers. Experimental results con rm that input decimation combining improves generalization performance.