ConceptAI & ML
adaLN-zero
An initialization strategy for DiT blocks where scale, shift, and gating parameters are initialized to zero to represent identity functions initially.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Adaptive layer normalization1ResNet initialization1
Lecturers
1Connections
2The lectures that define it or assume it, and the concepts next to it.