add a binary mask to do element-wise multiplication on weights in the fully-connected layer

[Shawn Fang]

Hi, everyone!

I'm trying to add a binary mask on the weights in the fully-connected layer. The value of the mask is fixed and has nothing to do with training. Basically, the weight matrix will multiply the the mask elementally-wise before it multiply each neuron. Can anyone tell me how to write this in the caffe prototxt ? Or is there a way to solve this? 

Thanks,

Shawn

[Przemek D]

There is an Eltwise layer that performs an element-wise operation of your choice (PROD (product), SUM or MAX) between its bottoms. You still need to load your mask somehow, perhaps as a data layer? Or a python layer?

[Shawn Fang]

Hello, thanks for answering. Is there any tutorial on how to build a data/python layer ?