Open miguelmartin75 opened 7 years ago
I'm comparing your model to caffe's implementation of AlexNet, where they use L2 regularisation for the weights (not bias).
Just wondering, why you don't do the same?
I'm comparing your model to caffe's implementation of AlexNet, where they use L2 regularisation for the weights (not bias).
Just wondering, why you don't do the same?