AI Glossary Term
What does “Data Augmentation” mean?
Techniques that create modified training examples to improve model generalization.
How Data Augmentation works
Example
A team with only a few thousand photos of plant leaves flips, crops, and tilts each image to produce many variants. The model sees more variety during training and is less likely to memorize exact pictures.
Why it matters
Collecting and labeling data is expensive. Augmentation stretches a small dataset and often improves performance on new data.
Common questions
Does augmentation work for text?
Yes, with methods such as swapping synonyms or paraphrasing, though it is trickier than for images because small changes can alter meaning.
Can augmentation hurt?
Yes, if the changes create unrealistic examples or alter the label, for instance flipping an image of the letter b into a d.
Is it the same as synthetic data?
Related but different. Augmentation modifies real examples, while synthetic data can be generated from scratch.