Multi Adaptive Hybrid Networks (MAHNet): Ensemble Learning in Convolutional Neural Network

Abstract

Convolutional neural networks are very successful in object classification and detection, thanks to constantly evolving neural network architectures. Although these architectures are quite successful, hybrid models can also be a better solution to different problems by combining with different architectures. In this study, it is aimed to create a hybrid model with different architectures. However, module selection is adaptive rather than predetermined. In order to achieve this, the new selection layer is selected from modules of different architectures trained in parallel with the ensemble learning perspective. The coefficient created for each module is used as a boosting ensemble learning. However, unlike boosting, module selection is made according to these weights.