search

XtractOpen / Meganet.jl

A fresh approach to deep learning written in Julia
http://www.xtract.ai/
Other
14 stars 9 forks source link

Replace getAverageMatrix with MKL convolution? #69

Open klensink opened 6 years ago

klensink commented 6 years ago

Could we replace the average pooling that is currently being done by the kronecker product from getAverageMatrix with a strided 2x2 convolution where all the weights are 0.25?

The MKL package is proving to be really fast so I think we should use it wherever we can

klensink commented 6 years ago

70 makes changes to the Connector that will make it possible to use a function instead of an array

lruthotto commented 6 years ago

Sure. This is a good idea!

Sent from my iPad

On Mar 6, 2018, at 2:38 AM, Keegan Lensink notifications@github.com wrote:

70 makes changes to the Connector that will make it possible to use a function instead of an array

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

klensink commented 6 years ago

It seems that MKL actually has implemented versions of min/max/average pooling and batch/TV norm. I think it would be interesting to give these a shot and see how much faster they are.

https://software.intel.com/en-us/mkl-developer-reference-c-deep-neural-network-functions