Neural α-feature detector for feature detection and generalization

We propose a neural α-feature detector used to extract a small number of main or essential features in input patterns. Features can be detected by controlling α-entropy for α-feature detectors. The α-entropy is defined by the difference between Renyi entropy and Shannon entropy. The α-entropy controller aims to maximize information contained in a few important α-feature detectors, while information for all other feature detectors is minimized. Thus, the α-entropy controller can maximize and simultaneously minimize information. The neural α-feature detector was applied to the inference of consonant cluster formation. Experimental results confirmed that by controlling α-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 α-entropy