What is data augmentation and its purpose in ML?

• A. Generating transformed copies of data to increase data diversity and improve generalization.
• B. Reducing data diversity to speed up training.
• C. Normalizing features to zero mean and unit variance.
• D. Regularizing the model with dropout.

Answer: (C)

Data augmentation is the practice of creating additional training examples by applying transformations to existing data. The goal is to expand the training set with diverse, yet label-preserving, variants so the model learns more robust patterns and generalizes better to unseen data. In image tasks, common transformations include rotations, flips, crops, and color adjustments; in text, paraphrasing or synonym replacement can be used; in audio, slight tempo changes or adding noise are typical. The key is that these edits reflect plausible variations the model might encounter in the real world, so the label remains correct. This approach directly increases data diversity and helps reduce overfitting by exposing the model to a broader range of inputs. It’s different from feature normalization, which scales inputs for optimization, and from dropout, which is a regularization technique applied during training to prevent over-reliance on particular neurons.