Broad intuition · Study map 11 of 85
Generalization: learning the pattern, not the answer sheet
Suggested preparation: Clustering, compression, and anomalies
Picture it
A flexible model can memorize every training example. Regularization makes some solutions less attractive so the model has a better chance of working on new examples. Validation is the rehearsal; the test set is the final check.
What to understand
- Train/validation/test splits; leakage; bias and variance
- Weight decay, dropout, augmentation, early stopping
Follow the references
1 · Course
Google · Machine Learning Crash Course ↗Study overfitting and data preparation; look for how information can leak across a split.
You understand it when…
A model gets perfect training accuracy and poor new-user accuracy. Which measurement and intervention would you try first?
This is a study map. Follow the reference for the full explanation; use the question to test your mental model. Browse the reference shelf.