Neural α-feature detector for feature detection and generalization

In this paper, we propose a neural alpha-feature detector used to extract a small number of main or essential features in input patterns. Features can be detected by controlling alpha-entropy for alpha-feature detectors. The alpha-entropy is defined by the difference between Renyi entropy and Shannon entropy. The alpha-entropy controller aims to maximize information contained in a few important alpha-feature detectors, while information for all other feature detectors is minimized. Thus, the alpha-entropy controller can maximize and simultaneously minimize information. The neural alpha-feature detector was applied to the inference of the consonant cluster formation. Experimental results confirmed that by controlling alpha-entropy a small number of principal features can be detected, which can intuitively be interpreted. In addition, we could see that generalization performance is improved by minimizing alpha-entropy.