Evolution Strategies for Deep Neural Network Models Design.
Petra Vidnerová, Roman Neruda · ASEP · 2017
Deep neural networks have become the state-of art methods in many fields of machine learning recently. Still, there is no easy way how to choose a network architecture which can significantly influence the network performance. This work is a step towards an automatic architecture design. We propose an algorithm for an optimization of a network architecture based on evolution strategies. The algorithm is inspired by and designed directly for the Keras library [3] which is one of the most common implementations of deep neural networks. The proposed algorithm is tested on MNIST data set and the prediction of air pollution based on sensor measurements, and it is compared to several fixed architectures and support vector regression.