A Gaussian Process-Based Incremental Neural Network for Online Regression.

This paper proposes a Gaussian process-based incremental neural network algorithm to handle the online regression problem. It can extract prototypes by an incremental neural network, where 1) Gaussian process approximations are adopted to update the threshold regions and the posterior distribution of the dependent variable at the weight vectors of nodes and 2) the optimal bandwidth matrix is derived for adapting to network structure. Besides, we discuss some properties of the proposed approach, and the experimental results show that our approach achieves remarkable accuracy improvement in extracting prototypes for online regression on noisy data.