How does a computer identify pictures of cats? Or drive a car? These are jobs for deep learning. Deep learning is a specialization of machine learning for perceptual tasks. If this sounds hard, it’s because it is! The Python community has contributed two libraries TensorFlow and Keras to help.
This session is a survey of deep learning fundamentals with an introduction to more advanced topics. A majority of the first half is spent on building a conceptual neural network, which is what makes deep learning work. No code is shown, just the facts. After that, the TensorFlow library will be discussed with a practical code example, but not at length. The focus of the session, and the entire second half, is Keras. This is a higher-level library that avoids the complexity of TensorFlow thus making the power of deep learning accessible. And there is no performance tradeoff either as Keras is built on top of TensorFlow. This will segue into more advanced topics including convolutional neural networks, recurrent neural networks and a brief introduction to GANs and reinforcement learning.