On 2017-03-02 06:13, jmh530 wrote:
On Wednesday, 1 March 2017 at 21:48:27 UTC, Faux Amis wrote:
Question, do you know of any (plans to make a) deep learning framework
in D? (like: http://caffe.berkeleyvision.org)
I only recall a previous discussion on this thread
As noted on that thread, it's a lot of work, but if you're interested in
it there's no reason you can't help make contributions of some kind or
another. Rather than think about the whole library, you could think
about what would be precursors to a good machine learning library.
Integration between Mir GLAS and Dcompute is pretty obvious as machine
learning libraries are pretty computationally intensive. Along the same
lines, an autodiff library would be very helpful. I would expect that D
would provide a good framework for building one, especially given Ilya's
success with GLAS. Machine learning libraries also make heavy use of
graphs. I see that there is a Dgraph, but the dub page says it is still
experimental. I'm sure there's a bunch of other things that I haven't
even thought of.
Thanks for the link!
I agree with most what is said in the thread. Especially that it would
take lots of work, but that D might be a good fit for the task.
Currently working an Quiver container in UE4's C++ :(
But, autodiff seems like a nice project..