Regression Testing
Domain: Evaluation
🎯 Motivation
Re-testing to ensure new changes did not break existing functionality. Essential for model updates.
📋 Concrete Example
Model v2 scores 85% on new features but 70% on old benchmarks (was 88% in v1). Regression! v2 broke something. Fix before release.
🔗 Analogy
A student re-checking old topics after learning new ones - ‘Did I forget algebra while learning geometry?’ Regression testing ensures new learning does not break old knowledge.