Data augmentation as regularization in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is data augmentation in machine learning?

Data augmentation is a technique where we create new training examples by slightly changing existing data. This helps the model learn better by seeing more variety.

beginner

How does data augmentation act as a regularizer?

By adding varied versions of data, data augmentation prevents the model from memorizing exact examples. This reduces overfitting and helps the model generalize better to new data.

beginner

Name three common data augmentation techniques for images.

Common techniques include flipping images horizontally, rotating images slightly, and zooming in or out.

intermediate

In TensorFlow, which module helps apply data augmentation easily?

TensorFlow's Keras API provides layers like tf.keras.layers.RandomFlip, RandomRotation, and RandomZoom to apply data augmentation inside the model pipeline.

beginner

Why is data augmentation preferred over just collecting more data?

Collecting more data can be expensive and slow. Data augmentation creates new examples quickly by modifying existing data, saving time and resources while improving model robustness.

What is the main goal of data augmentation?

A Remove noisy data from the dataset

B Reduce the size of the training dataset

C Make the model run faster

D Increase training data variety to reduce overfitting

Data augmentation increases the variety of training data to help the model generalize better and reduce overfitting.

Which of these is NOT a typical image data augmentation technique?

A Random flipping

B Random deletion of pixels

C Random rotation

D Random zooming

Random deletion of pixels is not a common augmentation technique; typical ones include flipping, rotation, and zooming.

In TensorFlow, which layer would you use to flip images randomly during training?

A tf.keras.layers.Conv2D

B tf.keras.layers.Dense

C tf.keras.layers.RandomFlip

D tf.keras.layers.Dropout

tf.keras.layers.RandomFlip is designed to flip images randomly as a data augmentation step.

How does data augmentation help with model generalization?

A By exposing the model to varied data examples

B By increasing the model size

C By reducing the number of training epochs

D By making the model memorize training data

Data augmentation exposes the model to varied examples, helping it learn patterns that generalize beyond the training set.

Which statement is true about data augmentation?

A It artificially increases training data size

B It replaces the need for any regularization

C It can be applied only to image data

D It always decreases model accuracy

Data augmentation artificially increases the size and variety of training data, improving model robustness.

Explain how data augmentation works as a regularization technique in simple terms.

Think about how changing training examples helps the model not memorize.

List some common data augmentation methods for images and why they help.

Imagine how changing images slightly can teach the model to recognize objects better.