SECAI Core · phase 3 of 15
Data augmentation
A process of generating additional, artificial data that is similar to existing data, primarily for the purpose of training machine learning models and also often for use in computer vision processes.
The Explain card
- Plain English
- Data augmentation creates extra artificial examples that resemble the existing data, giving a model more variety to learn from. It is especially common in computer vision.
- Example
- A team training a document-tampering detector rotates, blurs and re-lights its image samples so the model copes with photos taken on bad phone cameras.
- Why it matters
- Augmentation improves robustness, including against some adversarial inputs, but it can also amplify a poisoned sample into many variants, spreading a backdoor through the whole dataset.
- Hook
- One photo becomes a hundred cousins, for good or bad.
Where it sits in the deck
Phase 3: Data Fundamentals: Types, Pipelines, and Preparation
Models are only as good as their data — understand the raw material and the engineering that shapes it before covering how it flows or breaks.
