I can't seem to find a C++ based ML/AI framework that implements a wide variety of neural network algorithms. I've used Encog for these purposes when working in Java, but I don't see anything that's similar, functionality-wise, in C++. The closest I've seen is FANN, but it lacks some stuff, LMA & annealing for example.
EDIT: The best alternative I've found is Shark, but as I said, it's still lacking and has only the more commonly used features, no LMA, annealing or PSO or anything of that level.
Ok, so after a few days of on and off searching, I think I can pretty safely say that there is no such thing. None of the existing libs provide anything even close to the level of sophistication of Encog, and since one of the main reasons I wanted it in C++ was the fact that I kind of need QT integration, I suppose I'll just end up using Encog and Jambi.
Anyway, to those who stumble upon this somewhat later on / from Google, there's basically three frameworks/libraries that do offer somewhat reasonable features, and they should really be enough if you don't want LMA, PSO, annealing or any of that more exotic stuff:
That's pretty much it. The rest is immature/dead or just really too lacking functionality-wise to mention. If I ever decide to roll my own project (quite possible, as I pretty much need it for work), I'll be sure to update this place with a link.