Hi all

Been struggling to write this theano function for a while now...

What I want to do is outlined in my question i posted 
here: 
http://stackoverflow.com/questions/43477193/theano-convolving-every-channel-of-tensor-a-with-every-channel-of-tensor-b
 
(hope cross posting is OK - if not slap me on the wrist and I'll remove it)

Example: TensorA (1, 40, 28, 28) (representing the visible units of my RBM 
- 40 channel input); TensorB (1, 100, 19, 19) (representing the hidden 
units of my RBM - implies 100 filters are being learned)

For each channel in TensorA I want to convolve with a single channel from 
TensorB and keep the result, so essentially this would end up with 
ResultTensor(100, 40, 10, 10) where dim[0] = output channel, dim[1] = input 
channel and dim[2] = r amd dim[3] = c

The plain old python approach is really simple, but really slow.

Any advice? 

As i mention in the post, it would be really easy to accomplish with 
theano.tensor.signal.conv.conv2d(tensorA[0], 
tensorB[0]) but unfortunately theano.tensor.signal.conv.conv2d does not 
work on a GPU back end...

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to