Data augmentation as regularization in TensorFlow - ML Experiment: Train & Evaluate

Experiment - Data augmentation as regularization

Problem:Train a neural network to classify images from the CIFAR-10 dataset. The current model achieves 95% training accuracy but only 70% validation accuracy.

Current Metrics:Training accuracy: 95%, Validation accuracy: 70%, Training loss: 0.15, Validation loss: 1.0

Issue:The model is overfitting: it performs very well on training data but poorly on validation data.

Your Task

Reduce overfitting by using data augmentation as a regularization technique to improve validation accuracy to at least 80% while keeping training accuracy below 90%.

You must keep the same model architecture.

You can only add data augmentation layers or techniques.

Do not change the optimizer or learning rate.

Hint 1

Hint 2

Hint 3

Solution

TensorFlow

import tensorflow as tf
from tensorflow.keras import layers, models

# Load CIFAR-10 dataset
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()

# Normalize pixel values
x_train, x_test = x_train / 255.0, x_test / 255.0

# Define data augmentation pipeline
data_augmentation = tf.keras.Sequential([
    layers.RandomFlip('horizontal'),
    layers.RandomRotation(0.1),
    layers.RandomZoom(0.1),
])

# Define the model architecture (same as before)
inputs = layers.Input(shape=(32, 32, 3))

# Apply data augmentation only during training
x = data_augmentation(inputs)

x = layers.Conv2D(32, (3, 3), activation='relu')(x)
x = layers.MaxPooling2D((2, 2))(x)
x = layers.Conv2D(64, (3, 3), activation='relu')(x)
x = layers.MaxPooling2D((2, 2))(x)
x = layers.Conv2D(64, (3, 3), activation='relu')(x)

x = layers.Flatten()(x)
x = layers.Dense(64, activation='relu')(x)
outputs = layers.Dense(10, activation='softmax')(x)

model = models.Model(inputs=inputs, outputs=outputs)

# Compile the model
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

# Train the model with validation
history = model.fit(x_train, y_train, epochs=20, batch_size=64, validation_data=(x_test, y_test))

Added a data augmentation pipeline using RandomFlip, RandomRotation, and RandomZoom layers.

Inserted the augmentation pipeline as the first layer in the model to augment training images on the fly.

Kept the original model architecture and optimizer unchanged.

Results Interpretation

Before augmentation: Training accuracy was 95%, validation accuracy was 70%, showing strong overfitting.

After augmentation: Training accuracy decreased to 88%, validation accuracy improved to 82%, and validation loss decreased, indicating better generalization.

Data augmentation acts as a regularizer by increasing training data diversity, which helps the model generalize better and reduces overfitting.

Bonus Experiment

Try adding dropout layers after dense layers to further reduce overfitting and compare results.

💡 Hint

Add layers.Dropout(0.5) after the dense layer before the output layer and retrain the model.

Practice

1. What is the main purpose of data augmentation in TensorFlow models?

easy

A. To increase the size and variety of training data to improve model generalization

B. To reduce the number of layers in the model

C. To speed up the training process by using smaller batches

D. To convert images to grayscale before training

Solution

1. Step 1: Understand data augmentation concept

   Data augmentation creates new training examples by slightly changing existing data, which helps the model see more variety.

2. Step 2: Connect augmentation to model performance

   This variety helps the model learn general patterns instead of memorizing, improving its ability to work on new data.

3. Final Answer:

   To increase the size and variety of training data to improve model generalization -> Option A

4. Quick Check:

   Data augmentation = improve generalization [OK]

Hint: Augmentation adds data variety to help model learn better [OK]

Common Mistakes:

• Thinking augmentation reduces model size
• Confusing augmentation with batch size changes
• Assuming augmentation changes image color only

2. Which of the following is the correct way to add random horizontal flipping to a TensorFlow model using Keras preprocessing layers?

easy

A. tf.keras.layers.RandomFlip('vertical')

B. tf.keras.layers.RandomRotation(0.5)

C. tf.keras.layers.RandomFlip('horizontal')

D. tf.keras.layers.RandomZoom(0.2)

Solution

1. Step 1: Identify the correct augmentation layer for horizontal flip

   TensorFlow provides RandomFlip with options 'horizontal', 'vertical', or 'horizontal_and_vertical'. For horizontal flip, use 'horizontal'.

2. Step 2: Match the syntax

   The correct syntax is tf.keras.layers.RandomFlip('horizontal'). Other options apply different augmentations.

3. Final Answer:

   tf.keras.layers.RandomFlip('horizontal') -> Option C

4. Quick Check:

   RandomFlip('horizontal') = horizontal flip [OK]

Hint: RandomFlip('horizontal') flips images left-right [OK]

Common Mistakes:

• Using 'vertical' instead of 'horizontal' for horizontal flip
• Confusing RandomFlip with RandomRotation
• Using RandomZoom for flipping

3. Consider this TensorFlow code snippet for data augmentation:

import tensorflow as tf
aug_layer = tf.keras.Sequential([
  tf.keras.layers.RandomFlip('horizontal'),
  tf.keras.layers.RandomRotation(0.1)
])

input_image = tf.ones((1, 100, 100, 3))
augmented_image = aug_layer(input_image)
print(augmented_image.shape)

What will be the printed output?

medium

A. (1, 100, 100, 3)

B. (100, 100, 3)

C. (1, 100, 100)

D. (100, 3)

Solution

1. Step 1: Understand input and augmentation output shapes

   The input image has shape (1, 100, 100, 3), where 1 is batch size, 100x100 is image size, and 3 is color channels.

2. Step 2: Check if augmentation changes shape

   RandomFlip and RandomRotation only change pixel values, not shape. So output shape remains the same as input.

3. Final Answer:

   (1, 100, 100, 3) -> Option A

4. Quick Check:

   Augmentation keeps shape same = (1, 100, 100, 3) [OK]

Hint: Augmentation changes pixels, not shape [OK]

Common Mistakes:

• Assuming augmentation changes image dimensions
• Ignoring batch dimension in shape
• Confusing channel dimension with batch

4. You wrote this code to add random rotation to your model input:

aug_layer = tf.keras.layers.RandomRotation(1.5)
augmented = aug_layer(input_images)

But you get an error: "ValueError: Rotation factor must be between -1 and 1". What is the likely cause?

medium

A. The rotation factor 1.5 is valid; error is from input_images shape

B. Rotation factor 1.5 is out of allowed range; it must be between -1 and 1

C. Rotation factor 1.5 is interpreted as 50 radians, which is too large

D. Rotation factor must be a tuple, not a single float

Solution

1. Step 1: Check allowed range for RandomRotation factor

   RandomRotation expects a float between -1 and 1 representing the fraction of 2π radians.

2. Step 2: Identify the error cause

   The factor 1.5 > 1 triggers the ValueError since it is outside the allowed range [-1, 1].

3. Final Answer:

   Rotation factor 1.5 is out of allowed range; it must be between -1 and 1 -> Option B

4. Quick Check:

   Rotation factor must be between -1 and 1 [OK]

Hint: Rotation factor must be between -1 and 1 [OK]

Common Mistakes:

• Believing the rotation factor 1.5 is valid
• Confusing radians with fraction factor
• Passing tuple instead of float

5. You want to add data augmentation to your TensorFlow model to reduce overfitting. Which of these combined augmentation layers is best to add inside your model for regularization?

hard

A. tf.keras.Sequential([tf.keras.layers.RandomTranslation(0.5, 0.5), tf.keras.layers.RandomRotation(1.5)])

B. tf.keras.Sequential([tf.keras.layers.RandomContrast(0.5), tf.keras.layers.RandomBrightness(0.5)])

C. tf.keras.Sequential([tf.keras.layers.RandomCrop(50, 50), tf.keras.layers.RandomFlip('vertical')])

D. tf.keras.Sequential([tf.keras.layers.RandomFlip('horizontal'), tf.keras.layers.RandomRotation(0.2), tf.keras.layers.RandomZoom(0.1)])

Solution

1. Step 1: Evaluate augmentation types for regularization

   Good augmentations for regularization include flips, small rotations, and zooms to add variety without distorting data too much.

2. Step 2: Check each option for correctness and validity

   tf.keras.Sequential([tf.keras.layers.RandomFlip('horizontal'), tf.keras.layers.RandomRotation(0.2), tf.keras.layers.RandomZoom(0.1)]) combines horizontal flip, small rotation (0.2), and small zoom (0.1), all valid and helpful. tf.keras.Sequential([tf.keras.layers.RandomContrast(0.5), tf.keras.layers.RandomBrightness(0.5)]) uses RandomContrast and RandomBrightness, which are less common and may distort colors too much. tf.keras.Sequential([tf.keras.layers.RandomCrop(50, 50), tf.keras.layers.RandomFlip('vertical')]) uses RandomCrop(50, 50) which is valid syntax if input is larger. Vertical flip alone is less common. tf.keras.Sequential([tf.keras.layers.RandomTranslation(0.5, 0.5), tf.keras.layers.RandomRotation(1.5)]) uses RandomRotation(1.5) which is invalid (rotation factor must be ≤1).

3. Final Answer:

   tf.keras.Sequential([tf.keras.layers.RandomFlip('horizontal'), tf.keras.layers.RandomRotation(0.2), tf.keras.layers.RandomZoom(0.1)]) -> Option D

4. Quick Check:

   Combine valid flips, rotations, zooms for best augmentation [OK]

Hint: Use small flips, rotations, zooms for effective augmentation [OK]

Common Mistakes:

• Using invalid rotation factors >1
• Choosing only color augmentations
• Applying too large crops or translations