Hi,
I’m currently working with auto-nng and a bunch of (whole numbered) input numbers which are related to some output numbers. Here auto-nng is trained with a set of these numbers and automatically adjusts the size and layers of its neuronal network to fit to the training data. After this training/adjustment is done, auto-nng works with the neuronal network generated this way.
Since I have more and more input data, auto-nng (which is a single-thread application only) becomes too slow and I want to switch over to a CUDA-based system (GeForce graphics card available and all Linux drivers installed, SETI is already number-crunching on the graphics cards).
I understand cuDNN is a library and not a ready-to-use software, so I’m looking for a starting point:
- is there an application like auto-nng already available which self-adjusts its own neuronal network iteratively?
- if not, are there some example applications available somewhere which demonstrate simple usage with some training data consisting of plain numbers?
- any other idea where to start to get into cuDNN?
Thanks :-)