The question I get asked most is, "Should I use TensorFlow or PyTorch for deep learning?" The YouTube video above is my 26-minute response, detailing the pros and cons of both libraries.

The talk begins with a survey of the primary families of deep learning approaches: Convolutional Neural Networks, Recurrent Neural Networks, Generative Adversarial Networks, and Deep Reinforcement Learning. (Thanks to the Belgian artist Aglaé Bassens for the stunning illustrations that feature throughout this section of the slide deck.)

If you're already familiar with deep learning, you can skip ahead to the 12:24 mark. Via interactive demos, the meat of the talk appraises the two leading deep learning libraries: TensorFlow and PyTorch. With respect to both model development and production deployment, the strengths and weaknesses of the two libraries are detailed with code examples from reference Jupyter notebooks. There's a particular focus on the TensorFlow 2 release that integrates the easy-to-use, high-level Keras API as a formal module within the library.

Thanks to Jared Lander and Amada Echeverría for hosting me at the terrifically well-executed virtual version of this year's New York R Conference, where I provided this lecture. (Note that the content I covered is equally relevant to folks working primarily in Python.)