ScholarGate
Assistent
Machine learning

Knowledge Distillation

Knowledge Distillation er en modelkomprimeringsteknik, introduceret af Geoffrey Hinton og kolleger i 2015, der træner en lille studentermodel ved hjælp af soft-label-output fra en stor teachermodel. Destillerede modeller som DistilBERT og TinyBERT opnår ca. 97% af den større models ydeevne, mens de kører langt hurtigere.

Åbn i MethodMindSnartVideoSnartDownload slides

Læs hele metoden

Kun for medlemmer

Log ind med en gratis konto for at læse dette afsnit.

Log ind

Method map

The neighbourhood of related methods — select a node to explore.

+3 more

Kilder

  1. Hinton, G., Vinyals, O. & Dean, J. (2015). Distilling the Knowledge in a Neural Network. NeurIPS Deep Learning Workshop. link
  2. Sanh, V., Debut, L., Chaumond, J. & Wolf, T. (2019). DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. arXiv:1910.01108. link

Sådan citerer du denne side

ScholarGate. (2026, June 1). Knowledge Distillation (Teacher–Student Model Compression). ScholarGate. https://scholargate.app/da/deep-learning/knowledge-distillation

Which method?

Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.

Compare side by side

Refereret af

ScholarGateKnowledge Distillation (Knowledge Distillation (Teacher–Student Model Compression)). Hentet 2026-06-15 fra https://scholargate.app/da/deep-learning/knowledge-distillation · Datasæt: https://doi.org/10.5281/zenodo.20539026