Data augmentation
Expanding a dataset by making tweaked copies of what you already have.
Think of it like
Like practicing a recipe with slight variations each time — different pans, small substitutions — so you can cook it in any kitchen.
Example
A vision model’s training images get flipped, cropped, and recolored, multiplying the dataset and teaching robustness.
How it actually works
Augmentation transforms existing examples in label-preserving ways — rotations for images, paraphrases or back-translation for text — to increase variety without collecting new data. It reduces overfitting and improves generalization by exposing the model to more surface variation. The catch: transformations that change the true label, or that don’t reflect real-world variation, teach the wrong lesson.
For product teams
A cheap robustness boost from data you already own — squeeze more learning from every example.
For engineers
Label-preserving transformations that expand data variety; curbs overfitting, improves generalization.
Related
- Overfitting — Reduces overfitting.
- Generalization — Improves generalization.
- Synthetic data — Overlaps with synthetic data.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome