Pain Point #6: Prioritizing high-level libraries for CNNs

There are many libraries available to perform convolutional neural networks. Some of them are considered low-level such as TensorFlow, where much of the configuration and setup requires extensive coding. This can be considered a major pain point for an inexperienced developer. There are other libraries, such as Keras, that are high-level frameworks built on top of libraries such as TensorFlow. These libraries require much less code to get up and running with building a convolutional neural network. Often times developers getting started with building a neural network will try and implement a model with TensorFlow and run into several issues along the way. This section will propose initially building a convolutional neural network with Keras instead to predict the hand-written images from the MNIST dataset.

...