SiamJEPA: On the Role of Siamese Student Encoders in JEPA

SiamJEPA: On the Role of Siamese Student Encoders in JEPA

Siamese student encoders in JEPA models improve representation separability and training efficiency through effective regularization, outperforming single-encoder variants and MAE…

Hugging Face · Daily Papers ·Makoto Yamada ·

Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.

Autores: Makoto Yamada

  • 0 upvotes da comunidade
  • Temas: Joint Embedding Predictive Architectures, JEPAs, masked autoencoders, Siamese encoders, exponential moving average, teacher network

Resumo

Resumo original (em inglês), extraído do paper:

Siamese student encoders in JEPA models improve representation separability and training efficiency through effective regularization, outperforming single-encoder variants and MAE under limited training budgets.

Onde ler

compartilhar: