Use of batch normalization in assignment exercise-1?

[Aryan06]

I was doing the programming assignment 2 of week 1 “Convolution_model_Application”. In exercise 1 while building sequential model we are using batch normalization for axis 3. As per my understanding axis 3 is for channels.
Q1) Why is batch normalization being used specifically for axis 3?

Also after running the command happy_model.summary(), it was shown that the params for batch normalization were 128.
Q2) How did batch normalization resulted in creating parameters?

. Here 128 is specifying the parameters.

[rmwkwok]

Hello @Aryan06,

@Aryan06

Q2) How did batch normalization resulted in creating parameters?

Please inspect the output of the following code that will show you some names of the parameters

import tensorflow as tf
X = np.random.rand(10,64,64,32)
l = tf.keras.layers.BatchNormalization()
_ = l(X)
print(l.weights)

then, read the documentation for the explanation of those parameters.

@Aryan06

Q1) Why is batch normalization being used specifically for axis 3?

Because this will give us a set of normalization parameters per channel (or per feature (map))

Raymond