Here are a few that I could think of which we already have in the ecosystem:

1. Classical signal processing, like what NxSignal implements
2. General numerical calculus, not only matrix math – although @sodapopcan has an example. Integrator is such an example, and you could even use that one to finish off this circuit sim I started porting in early Nx days.
3. Scholar also has non-ML stuff like interpolation algorithms, and some things which you could say are pre-ML statistics like Linear and Logistic Regression. The latter ones are a bit on the frontier because ML in the end is just statistics with another name.

These are a few areas that came to mind, but my main point is that ML to me is more related to statistics and stochastic processing, but you can use Nx for most use cases that require number crunching.

It’s handy for plain ol’ matrix math (I say that like I’m an expert—I’m not ). Image uses it in some non-ML ways, like for warp perspective, though it does use it for ML stuff as well.