Yes, they solve completely different problems.

"Mobile-dev" is more or less application development for a target client, possibly with a client-server relationship. It may incorporate the products of a ML system, or any number of other systems, but is itself the same application development model we've had for 30 years, but with a mobile client.

Machine Learning is applied computational statistics. It and its close cousin "data science" have become incredibly hyped the last few years. I suppose it's a matter of taste, but my view is that applying simple probability to a problem (e.g. "recommending" based on picking the thing most frequently voted up by other users and adjusting the recommendation as data comes in) isn't "machine learning" or "data science", but I've seen it called such by candidates I've interviewed and even one colleague. It's not that simple, and trivializing the terms just contributes to the hype.

Sorry, didn't make it clear, was wondering given all the difference why would one prefer one over the other? Seem like 'in the field' experience is not going to be very much different. Appears to me mobile can be even more fun.

So, i'm pretty much a dilettante, so take this with some salt. I have a pretty good handle on baby level ML stuff like PCA, k-means, decision tree forests. pretty much the stuff in Mackay and Russel&Norvig.

It really depends on what you mean by 'mobile developer'. If you're the kind of guy that can open up the RFC for SNMP and turn that spec into code, you're probably in decent shape. You can take abstract, but very specific, description and turn it into reliable code.

Even better would be to take an image processing paper and turn that into code. Maybe something like seam carving to delete stuff from photos. This is better than an RFC, because papers skip steps that are 'obvious' to readers. Video stabilization, AR, there are a bunch of applications that have more intense math.

At the end of the day, it's just "finding the minimum" of some function. (like programming is just zeros and ones, easy right?) Each paper is some trick to descend a gradient in a new way, or hop over a local minima or avoid overfitting.

You don't really need to understand a whole lot. You can patternmatch math syntax and retype that as code. if you make an error, debugging will suck in that case. But really, you can do a lot of damage without knowing a whole lot. Some linear algebra, some diff eq. You don't have to be super fast at it, there's nothing wrong with taking a couple of days where a math grad student knows off the top of their head. The goal is simply faithful translation.

So if a developer just sticking buttons on a panel, that mobile developer probably wouldn't make it. But there are a bunch of mobile developers that could make the switch.

Few subfields of CS are as hyped as ML.

Hype aside, it's still pretty useful and also pretty different from mobile development.