Category coding with neural network application

Abstract

In many applications of neural network, it is common to introduce huge amounts of input categorical features, as well as output labels. However, since the required network size should have rapid growth with respect to the dimensions of input and output space, there exists huge cost in both computation and memory resources. In this paper, we present a novel method called category coding (CC), where the design philosophy follows the principle of minimal collision to reduce the input and output dimension effectively. In addition, we introduce three types of category coding based on different Euclidean domains. Experimental results show that all three proposed methods outperform the existing state-of-the-art coding methods, such as standard cut-off and error-correct output coding (ECOC) methods.