Machine Learning Course - Transfer Learning

Finetuning - Replacing the Classification Head

COVID-19 Pulmonary CT Case Study - Interactive Simulation

Layer freeze control - click any row to toggle

Frozen layers (∇ = 0) act as fixed feature extractors - weights do not update during backprop.
Unfrozen layers receive gradient updates. For small medical datasets like pulmonary CT, freezing most layers prevents overfitting. Changes here carry forward into Phase 04.
Frozen (∇ = 0)
Unfrozen (∇ ≠ 0)
Training - new head
Removed