Co-Trained Random Vector Functional Link Network

In this paper, we propose ensemble of random vector functional link network known as co-trained random vector functional link network (coRVFL). Random vector functional link network solves the optimization problem via closed form solution and hence avoids the problems of slow convergence and local minima problems. The proposed coRVFL trains two RVFL models jointly such that each RVFL model is constructed with different feature projection matrix and hence, shows better generalization performance. We use randomly projected features and sparse-l(1) norm autoencoder based features to train the proposed coRVFL model. Experimental results show that the proposed coRVFL is performing better in comparison with the baseline models. Furthermore, statistical analysis reveals that the proposed coRVFL model performs statistically better than the baseline approaches.