SharpNEAT v2.3.0

14 views
Skip to first unread message

Colin Green

unread,
Apr 29, 2017, 4:43:54 PM4/29/17
to ne...@yahoogroups.com, shar...@googlegroups.com

Colin Green

unread,
May 21, 2017, 3:55:28 PM5/21/17
to ne...@yahoogroups.com, shar...@googlegroups.com
On 13 May 2017 at 16:49, Milind Dalvi milind....@yahoo.in [neat]
<ne...@yahoogroups.com> wrote:
>
>
> For C# it seems
> Alea GPU will do the job. The library does use licence for Enterprise level
> GPU like NVDIA Tesla and Grid but its free for consumer level with GeForce
> and Quadro (more on licencing here)
>

Very interesting, thanks for the link.

I've added it the roadmap as something to investigate further. I won't
be getting around to this anytime soon though, but in the mean time
there are some performance improvements in the works around activation
functions, single precision weights, and some basic use of SIMD/vector
CPU instructions. All of that combined with the recent tweak to the
speciation algorithm could deliver something like a 10x speedup
overall just running on a CPU (with 256 bit vector registers), but
time will tell. (also note that Intel gen 8 CPUs will have 512 bit
vector register when they arrive)

Thanks again for the pointer.

Colin
Reply all
Reply to author
Forward
0 new messages