A survey of deep-learning frameworks

Abstract

Deep learning is a model of machine learning loosely based on our brain. Artificial neural network has been around since the 1950s, but recent advances in hardware like graphical processing units (GPU), software like cuDNN, TensorFlow, Torch, Caffe, Theano, Deeplearning4j, etc. and new training methods have made training artificial neural networks fast and easy. In this paper, we are comparing some of the deep learning frameworks on the basis of parameters like modeling capability, interfaces available, platforms supported, parallelizing techniques supported, availability of pre-trained models, community support and documentation quality.