EDEN: Evolutionary Deep Networks for Efficient Machine Learning

摘要: Deep neural networks continue to show improved performance with increasing depth, an encouraging trend that implies explosion in the possible permutations of network architectures and hyperparameters for which there is little intuitive guidance. To address this complexity, we propose Evolutionary DEep Networks (EDEN), a computationally efficient neuro-evolutionary algorithm interfaces any deep platform, such as TensorFlow. We EDEN evolves simple yet successful built from embedding, 1D 2D convolutional, max pooling fully connected layers along their hyperparameters. Evaluation across seven image sentiment classification datasets shows it reliably finds good -- three cases achieves state-of-the-art results even on single GPU, just 6-24 hours. Our study provides first attempt at applying neuro-evolution creation convolutional analysis including optimisation embedding layer.