Multi-task modular backpropagation for feature-based pattern classification

Abstract

Modular knowledge development in neural networks have the potential to feature robust decision given sudden changes in the environment or the data during real-time implementation. It can also provide a means to address robustness in decision making given certain features of the data are missing post training stage. In this paper, we present a multi-task modular backpropagation algorithm that features developmental learning where the training takes into account several groups of features that constitute the overall task. The proposed algorithm employs multi-task learning where knowledge from previously trained neural network modules are used to guide knowledge developmental in future modules. The results show that it is possible to implement a modular network without losing training or generalization performance.