deep-learningtraining
Regularization
Dropout, weight decay, early stopping, data augmentation — preventing overfitting in deep networks.
Уровни глубины
L0Intro~1ч
Knows overfitting and that regularisation reduces it; has heard of dropout and weight decay.
L1Basics~8ч
Applies L1/L2 weight decay, dropout, early stopping, and basic data augmentation.
L2Working~15ч
Diagnoses under/overfitting via learning curves; applies BatchNorm, MixUp, label smoothing, and stochastic depth.
L3Advanced~30ч
Understands PAC-Bayes bounds; applies sharpness-aware minimisation (SAM), R-Drop, and calibration.
L4Research~60ч
Contributes to implicit regularisation theory or generalisation bounds research.
Ресурсы
L1 — Basics
L2 — Working