On Thursday, 7 January 2016 at 00:34:45 UTC, Henry Gouk wrote:
On Tuesday, 5 January 2016 at 12:27:12 UTC, Ola Fosheim Grøstad
wrote:
2. integrate with numerical solutions like TensorFlow?
I'm currently working on something along the same lines as
TensorFlow -- I started work on this before TensorFlow was
announced. I haven't thought too much about writing a
distributed backend, but currently there are some reasonably
inefficient proof-of-concept CPU and CUDA backends.
It's still fairly early on in development, and I'm more of an
academic than an engineer. That said, I've managed to train
some modestly sized convolutional networks without trouble.
I'll open source it at some point, but I haven't got a
time-frame for that.
Interesting project.