A Novel Plastic Neural Model with Dendritic Computation for Classification Problems

Junkai Ji, Minhui Dong, Cheng Tang, Jiajun Zhao, Shuangbao Song

Abstract: This paper proposes a novel plastic neural model (PNM) at the single neuron level and a specified learning algorithm to train it. The dendritic structure of PNM presents its diversity to fulfill each particular task. During the training process, PNM divides the Euclidean space of the training instances into several appropriate hypercubes, which have the same dimensional number. And then, each hypercube is transformed into a corresponding dendritic branch in PNM. A suitable dendritic structure of PNM has been proved to have powerful computational capabilities to solve the classification problems. Both theoretical analysis and empirical study of the proposed model are demonstrated in this paper. Five benchmark problems are employed to verify the effectiveness of PNM in our experiment. The results have verified that PNM can provide competitive classification performance compared with several widely-used classifiers.