One thing I'd add though, is that physicists still have a healthy amount of skepticism on ML/NN. They don't like the "black box" approach, where the algorithm is regarded to "provide" the truth, but we're required to fully understand the math behind any implementation, as well as to be able to answer deeper questions on the results we're acquiring.
Background : I'm completing my master in Particle Physics, and have had the opportunity to work along these kinds of people. They have wrestled their entire careers with enormous amounts of data, create new solutions to problems the industry had not even formulated, and keep themselves on the forefront of Statistics and ML.
I feel that the things I've learned by trotting along grey-haired old-school Unix wizards could not be replicated by any courses, bootcamps or manuals.