Abstract

It has been proven that deeper Convolutional Neural Networks (CNN) can result in better accuracy in many problems, but this accuracy comes with a high computational cost. Also, input instances have not the same difficulty. As a solution for accuracy vs. computational cost dilemma, we introduce a new test-cost-sensitive method for convolution neural networks. This method trains a CNN with a set Based on the difficulty of the input instance, the expert branches decide to use a shallower part of the network or go deeper to the end. The expert branches learn to determine: is the current network prediction wrong and if the instance passed to deeper network layers it will generate the right output; if not, then the expert branches will stop the process of computation. The experimental results on the standard CIFAR-10 dataset indicate that in comparison with basic models, the proposed method can train models with lower test cost and competitive accuracy.