How to make the weight obtained by convolution2dlayer non-negative？

I am training a simple network. I need to constrain the weights of one layer of convolution2dlayer to be >0. For example, in keras, Layer weight constraints are implemented using tf.keras.constraints.NonNeg(). Can this function be realized in matlab? If there is no direct way, are there indirect constraints?

Answer 1

[Abhishek Gupta]

Hi,

As per my understanding, you want to constrain the weights of a convolution2DLayer in MATLAB. You can do so by accessing the weights of a layer as shown below: -

net.LW{i,j}

Referring to the following resources, which might help you in resolving the issue: -

• How to constrain the Neural Network output weights to fixed values
• MATLAB Documentation - Neural Network Subobject Properties - Layer Weights

[CHAO WANG]

Thank you very much for your answer. It's helpful.I will try to run it on my program.