Generation and Frame Characteristics of Predefined Evenly-Distributed Class Centroids for Pattern Classification

Predefined evenly-distributed class centroids (PEDCC) can be widely used in models and algorithms of pattern classification, such as CNN classifiers, classification autoencoders, clustering, semi-supervised learning, etc. Its basic idea is to predefine the centers, which are on unit hypersphere feature space, maximize inter-class distance. The previous method generating PEDCC uses an iterative algorithm based a charge model. generated centers will have some errors with theoretically points, generation time long. This paper takes advantage regular polyhedron high-dimensional space evenly distributed points n dimensional generate mathematically. Then, we discussed extended characteristics frames formed by PEDCC, meaningful conclusions obtained. Finally, effectiveness new related proved experiments. mathematical analysis experimental results this provide theoretical tool for using solve key problems field recognition, interpretable supervised/unsupervised incremental uncertainty so on.