Core ML Concepts · intermediate · concept 29 of 176
Data Augmentation
Stretching a dataset by transforming what you have: flips, crops, and color jitter for images, mixup and CutMix across examples, back-translation for text, SpecAugment for audio. A data-side regularizer that buys accuracy for free when labels are expensive, and standard practice in every vision pipeline.
Key terms
Flips & cropsMixup & CutMixBack-translationSpecAugmentLabel-preserving transforms
Learn these first
Where you meet it in the real world
Every vision model in production, low-data medical imaging, robust speech recognition
Videos
▶ C4W2L10 Data Augmentation ↗
DeepLearningAI · YouTube
Guides and articles
Courses, papers, and more