Batch normalization in TensorFlow - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to add batch normalization after a dense layer in TensorFlow.

TensorFlow

model = tf.keras.Sequential([
    tf.keras.layers.Dense(64, activation='relu'),
    tf.keras.layers.[1](),
    tf.keras.layers.Dense(10, activation='softmax')
])

Drag options to blanks, or click blank then click option'

ABatchNormalization

BDropout

CFlatten

DConv2D

Attempts:

3 left

💡 Hint

Common Mistakes

Using Dropout instead of BatchNormalization.

Forgetting to add batch normalization after the dense layer.

The BatchNormalization layer normalizes the outputs of the previous layer, helping training stability.

2fill in blank

medium

Complete the code to create a batch normalization layer with momentum 0.9.

TensorFlow

bn_layer = tf.keras.layers.BatchNormalization(momentum=[1])

Drag options to blanks, or click blank then click option'

A0.5

B0.1

C1.0

D0.9

Attempts:

3 left

💡 Hint

Common Mistakes

Setting momentum to 1.0 disables moving average updates.

Using too low momentum like 0.1 can make training unstable.

The momentum parameter controls the moving average update speed. 0.9 is a common default.

3fill in blank

hard

Fix the error in the batch normalization usage by completing the code.

TensorFlow

x = tf.keras.layers.Dense(128)(inputs)
x = tf.keras.layers.BatchNormalization[1](x)

Drag options to blanks, or click blank then click option'

A.apply

B()

C.call

D.fit

Attempts:

3 left

💡 Hint

Common Mistakes

Using .call or .apply instead of calling the layer as a function.

Forgetting parentheses causes a syntax error.

In TensorFlow, layers are called like functions with parentheses to apply them to inputs.

4fill in blank

hard

Fill both blanks to create a batch normalization layer that normalizes over the last axis and uses epsilon 1e-5.

TensorFlow

bn = tf.keras.layers.BatchNormalization(axis=[1], epsilon=[2])

Drag options to blanks, or click blank then click option'

A-1

B0

C1e-5

D1e-3

Attempts:

3 left

💡 Hint

Common Mistakes

Using axis 0 normalizes over batch dimension, which is incorrect.

Using too large epsilon like 1e-3 can reduce normalization effectiveness.

Axis -1 means normalize over the last axis (features). Epsilon 1e-5 avoids division by zero.

5fill in blank

hard

Fill all three blanks to create a model with batch normalization after each dense layer and dropout after the first batch norm.

TensorFlow

model = tf.keras.Sequential([
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.[1](),
    tf.keras.layers.[2](0.5),
    tf.keras.layers.Dense(64, activation='relu'),
    tf.keras.layers.[3](),
    tf.keras.layers.Dense(10, activation='softmax')
])

Drag options to blanks, or click blank then click option'

ABatchNormalization

BDropout

DFlatten

Attempts:

3 left

💡 Hint

Common Mistakes

Placing dropout before batch normalization.

Forgetting to add batch normalization after the second dense layer.

BatchNormalization layers normalize activations. Dropout randomly disables neurons to reduce overfitting.