Architectures · advanced · concept 96 of 311 · Updated Nov 4, 2026
Knowledge Distillation
Training a smaller 'student' model to mimic a larger 'teacher' model's behavior. The student learns from the teacher's soft outputs (probabilities), capturing knowledge that's lost with hard labels alone.
Key terms
Teacher modelStudent modelSoft labelsTemperature scaling
Learn these first
Videos
▶ Understanding Knowledge Distillation in Neural Sequence Generation ↗
Microsoft Research · YouTube
Guides and articles
Courses, papers, and more