Hyper-parameter optimization: toward Convolutional

Abstract

n a broad range of computer vision tasks, convolutional neu- ral networks (CNNs) are one of the most prominent tech- niques due to their outstanding performance. Yet it is not trivial to find the best performing network structure for a spe- cific application because it is often unclear how the network structure relates to the network accuracy. We propose an evo- lutionary algorithm-based framework to automatically opti- mize the CNN structure by means of hyper-parameters. Fur- ther, we extend our framework towards a joint optimization of a committee of CNNs to leverage specialization and coop- eration among the individual networks. Experimental results show a significant improvement over the state-of-the-art on the well-established MNIST dataset for hand-written digits recognition. Index Terms—