Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
6625 Discussions

Implementing a convolutional neural network using MKL

jean-vezina
Beginner
208 Views

Good evening,

After discovering that it is not possible to build MKL-DNN on Windows, I would like to know if there is an example code or some hints about implementing a convolutional neural network (CNN) using the primitives already implemented in the MKL library available with XE 2017 release 2?

Thanks in advance,

Jean Vezina

 

0 Kudos
4 Replies
Gennady_F_Intel
Moderator
208 Views

mkl-dnn-0.7 version is available from github. You may have a look at the source code and see the implementations details (e.x - ..\intel-Caffe\mkl-dnn-0.7\src\cpu\jit_avx2_1x1_conv_kernel_f32.cpp - you will show the AVX2 specific implementation ) 

jean-vezina
Beginner
208 Views

Thanks a lot! 

Regards,

Jean

 

SergeyKostrov
Valued Contributor II
208 Views
>>...some hints about implementing a convolutional neural network (CNN)... If you're at initial phase of your R&D I would recommend to look at a set of convolution functions of IPP. >>...After discovering that it is not possible to build MKL-DNN on Windows Developers of MKL-DNN overcomplicated simple matters and they've imposed too many constraints that prevent MKL-DNN to be widely accepted in an Open Source community.
jean-vezina
Beginner
208 Views

Thanks,

I will check the convolution functions of IPP.

FYI, I have also noticed that some other deep learning tools that call MKL such as the Intel versions of Caffe and Neon cannot also be compiled on Windows as CMAKE for Windows returns errors and Intel C++ finds errors in the source code when we try to compile manually. Apparently, one has to use GNU C++ (g++) to compile without errors but this is not acceptable in my case as the compiled code cannot be called by Intel Fortran or C++.

Regards,

Jean

 

Reply