Calculate number of training parameters using Neural Network Toolbox

Question

I have built a CNN using the Matlab Neural Network Toolbox. How can I calculate the total number of parameters used in my network?

For example, I can use the analyzeNetwork() from the support addon, and this can show me the number of learnable parameters per layer. Would it be correct to add these up?

layers = [
imageInputLayer([128 300 1])

convolution2dLayer([7 7], 32, 'Stride', [2 2], 'Padding', 1)
batchNormalizationLayer
reluLayer

convolution2dLayer([5 5], 32, 'Stride', [2 2], 'Padding', 1)
batchNormalizationLayer
reluLayer

convolution2dLayer([3 3], 64, 'Stride', [2 2],'Padding', 1)
batchNormalizationLayer
reluLayer
dropoutLayer(0.2)

convolution2dLayer([3 3], 64, 'Stride', [2 2],'Padding', 1)
batchNormalizationLayer
reluLayer
dropoutLayer(0.2)

convolution2dLayer([8 1], 1024, 'Stride', [1 1],'Padding', 0)
batchNormalizationLayer
reluLayer
dropoutLayer(0.2)

averagePooling2dLayer([1 12],'Stride', [1 1]);
dropoutLayer(0.2)

fullyConnectedLayer(4096)
batchNormalizationLayer
reluLayer

dropoutLayer(0.2)

fullyConnectedLayer(1251)
softmaxLayer
classificationLayer];

What are imageSize and numClasses? Please add this to your question. — Dev-iL
– Dev-iL, Commented Jun 17, 2018 at 15:43
The network analyzer shows that ReLU layers have 0 learnable parameters. This means it cannot be trusted... — Dev-iL
– Dev-iL, Commented Jun 17, 2018 at 16:44

Jonty800 · Accepted Answer · 2018-06-17 22:28:00Z

0

I used analyzeNetwork() from the support addon, and added up all of the learnable parameters, equalling 1.4M.

I then created the same network in matconvnet library, and used vl_simplenn_display to analyse the network. It also showed 1.4M parameters.

answered Jun 17, 2018 at 22:28

Jonty800

5221 gold badge7 silver badges21 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Calculate number of training parameters using Neural Network Toolbox

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related