Undefined Function in CNN

[TYS11]

Undefined sigm in cnnff.m at line 17 and line 37.
Undefined expand function in cnnbp.m at line 26

[wajihullahbaig]

Did you add the path to the folders of DeepLearnToolBox?
YOu have to tell matlab the path to the whole tool box

[TYS11]

Thanks, solved it.
I wonder why I get so many Kernels at second Convolution layer.
I wanted to have a 8 filters(aka weights) at both of the Conv layers.
So for First conv layer I set outputmaps to 8, for second layer I set outputmaps to 8 as well,
but the filters it produced is 64 at second Conv layer, instead of 8?

[wajihullahbaig]

I am afraid I did not understand the question clearly. You get many kernels because you are using a whole batch of images and push them through the convolution layer. I don;t understand your second question but still may be you did not understand maps correctly in convolution?
A division by 255 keeps the data in the range of [ 0 , 1] and this is normal practice in computer vision.

[wajihullahbaig]

And for a better understanding on CNN filters,maps etc here is a good link
http://cs231n.github.io/convolutional-networks/

[TYS11]

by the way, your pooling layer uses max pooling or mean pooling?

[wajihullahbaig]

Max pooling

[TYS11]

thanks, much obliged.

[TYS11]

One last thing, softmax classifier in the output layer?

[wajihullahbaig]

If you are talking about the CNN, its sigm() found here https://github.com/rasmusbergpalm/DeepLearnToolbox/blob/master/util/sigm.m
which is inside the feed forward module, https://github.com/rasmusbergpalm/DeepLearnToolbox/blob/master/CNN/cnnff.m

For simple neural network, the tests show softmax output layer. Check the 'tests' folder for a better understanding.