|
Anyone knows a reliable library (with source code) for convolutional networks? I've been developing my own code in matlab and I'm having problems implementing back propagation for max pooling and subsampling layers. Thanks. |
|
I have a pretty convoluted deep networks in Matlab project on Github. It is constantly changing, and often stuff will become deprecated, but you could look at it for some ideas. The project is: ML-Framework My 2D max pooling layer is the file: ML-Framework / Layer / HiddenLayer / CompositeHiddenLayerParts / HigherLayers / Conv2DMaxPoolingLayer.m You can email me at my address listed on Github if you have any specific questions. |
|
Have a look at this code: https://github.com/dustinstansbury/medal It has a lot of models implemented. |
|
You can check out Theano which uses Python. UFLDL has tutorials using MATLAB but you need to fill up the source code... maybe someone has put up their code online. |
|
There is a github deeplearning toolbox which might help. thanks but it doesn't have max pooling.
(Nov 21 '13 at 10:36)
rm9
but they do have subsampling in here: https://github.com/rasmusbergpalm/DeepLearnToolbox/blob/master/CNN/cnnff.m Maybe, you could try to change line 25 to calculate the max inside the region instead of just downsampling... His idea for the gradient of this step is here https://github.com/rasmusbergpalm/DeepLearnToolbox/blob/master/CNN/cnnbp.m at line 32.
(Feb 04 '14 at 08:45)
eder
|
