Suggestions about Starting a Phd in Machine Learning – Yay or Hell No?

Changing topic area for a PhD is hard work, but usually of significant value, because you have insights into each from the other. As such, the work provides a return. But it's hard work.

Will the PhD be of net positive value? In some fields, to have a PhD is of net negative value (although perhaps only initially) because potential employers will believe you have only book-learning experience, and are of no practical use.

Do you want to do a PhD? Why? It's hard work, extremely frustrating, and usually somewhat marginal. If you work on truly mainstream material there is always a chance someone will publish before you and potentially prevent you from getting the PhD.

So if you really want to do it, then go for it, but start by understanding your reasons and motivations. Certainly ML is a rich field, full of as-yet-undiscovered techniques and algorithms, and undoubtedly useful. But remember, a PhD is not the only way to get into a field - another path might be better. Don't be excessively tempted for the initially easy option.

If you get a full scholarship then probably, that's usually the case though. The PhD will give you prestige and credibility if you start your own business, and will open some doors when you are looking for a job (but also close others). It will also set you back a few years from your goals, and the amount of money you will get with a PhD won't be more than with a MS. Best perk of having a PhD is working in academia, which can be a great life style: much lower stress, decent compensation, lots of vacation time and recognition.