Comparison of Tensorflow and PyTorch in Convolutional Neural Network - based Applications

In this paper, we present a comparison between the PyTorch and TensorFlow environments, used in defining neural networks. The purpose is to find whether the choice of a library affects the overall performance of the system both during training and design. To do so, our approach involves analysis of the processes involved when creating a neural network, as well as taking measurements and monitoring its evolution over the epochs. Advantages and disadvantages are then extracted from the results and are then used to draw conclusions.