Using Docker as the infrastructure for your Machine Learning environment can reduce drastically the burden of installing and replicating it over and over, on every server and working station.
In my talk, I show how you can immediately switch to using Docker. Here are my slides: https://docs.google.com/presentation/d/1F_UWuloTQiTt2fe9KiUOjLgUQ076g5HmFYFySDFsiE8/edit?usp=sharing