Just remember, most distros have live usb stick distros so you can always try out a bunch before you decide on the right one for yourself.
For servers, these days I'd recommend Alpine on ARM architecture for a very good mix of high performance and having sane defaults set up so you can easily set up a reverse proxy, web server, etc.
No advantage, but Ubuntu is the most popular distro for regular users / tutorial customers. Ubuntu also has the widest availability of support resources, even though the information is often not Ubuntu-specific.
If you use a non-Ubuntu (or non-Debian-derived) distro, you'll need to do a little bit of package-name mapping to get the prerequisites installed. This is annoying but only has to be done once (take notes!).
The bigger problem I've had with ML libs is that they're very picky about version compatibilities. Once you settle on a set of working/compatible versions (libs, python, python pkgs), make some effort to preserve your sources. Package versions can get deleted from the official repos, be prepared to build from source, etc.