Fork me on GitHub
#uncomplicate
<
2017-06-29
>
whilo13:06:58

@jduhamel which methods do you plan to apply?

whilo13:06:36

I am currently looking into implementing a GAN and I might start with the tensorflow interface for now, because porting of python code will be much more straightforward.

whilo13:06:00

Many people in my group use pytorch though, because torch can dynamically alter the compute graph, while tensorflow compiles it statically.

whilo13:06:36

I think the uncomplicate libraries might allow that as well, so I am currently playing around a bit with both.

whilo13:06:00

A problem is that the automatic differentiation is not yet implemented and that would mean yet another deep learning framework.

whilo13:06:13

I don't have time for that in my master thesis probably.

whilo13:06:27

If others work on it as well, that might be different though.