Using mdCNNs for regression

Philip Egger

unread,

Jul 10, 2018, 11:28:27 AM7/10/18

to mdCNN: Multidimensional CNN library in Matlab

Hi everyone,

I was wondering whether it is possible to use this mdCNN library not just for classification, but regression as well. In other words, can my labels live in continuous space as well, or only in discrete space?

Thanks,

Philip

Message has been deleted

rayjesc...@gmail.com

unread,

Sep 7, 2018, 3:13:05 PM9/7/18

to mdCNN: Multidimensional CNN library in Matlab

I was also interested in attempting to use this for 3D convolution and regression but I don't totally understand your suggested modifications. Is there any chance you could clarify a bit more? Thanks!

On Wednesday, July 11, 2018 at 8:59:58 AM UTC-5, Hagay Garty wrote:

Hi, yes it's possible. you need to modify the 'Train' function not to pass a binary vector to backPropegate
set expectedOut to be continues instead of binary

zero / one vector to ba

Hagay Garty

unread,

Nov 25, 2018, 1:57:43 AM11/25/18

to mdCNN: Multidimensional CNN library in Matlab

In the new version, you can set the regression layer loss function to a user-defined function or use the built-in crossEnt / MSE functions.

On Tuesday, July 10, 2018 at 6:28:27 PM UTC+3, Philip Egger wrote:

Reply all

Reply to author

Forward