Nearest convex hull classification by using Lotka-Volterra recurrent neural networks

摘要

Distance-based classifiers have been applied to many multi-class classification problems. The nearest convex hull classifier (NCHC) is a useful distance-based classifier. It assigns a test sample to the class that has the closest convex hull. This paper proposes a new algorithm to implement the NCHC. Considering an alternative interpretation of NCHC, the distance from the test sample to the convex hull of the training data in a certain class can be thought of as the reconstruction error. We propose an algorithm that uses neural networks to implement the NCHC. Our experimental results show that NCHC using Lotka-Volterra recurrent neural networks outperforms other classifiers in a whole.