Re: [caffe-users] Meaning of diff variable in blobs and params

[Evan Shelhamer]

`diff` is the gradient w.r.t. the data held by the blob. It is the same size as `data` since it is its gradient, whether the blob holds parameters or layer outputs.

Evan Shelhamer

On Fri, Jul 17, 2015 at 7:46 PM, zzz <mr.zizh...@gmail.com> wrote:

Hi,

In data blobs and parameters blobs, both hold a data variable and diff variable. I am kind of confused by the meaning of diff. I found the the size of diff in data blobs and parameters blobs are all same with the data saved in each layer.

For example, in the input data layer, the data variable size is N*3*128*128. Then the diff is the same.

Could someone explains the meaning of the diff variables in each layer (especially in data blobs)? How does they contribute the backpropagation?

Thanks!

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/31874136-7c82-4994-8d5d-2873c1ee8bdf%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.