Optimizing Visual Generative Models via Distribution-wise Rewards

Optimizing Visual Generative Models via Distribution-wise Rewards

A novel reinforcement learning framework for visual generation uses distribution-wise rewards to improve image diversity and quality while addressing mode collapse and computationa…

Hugging Face · Daily Papers ·Ruihang Li, Mengde Xu · ·▲ 12 upvotes

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

Autores: Ruihang Li, Mengde Xu, Shuyang Gu, Leigang Qu, Fuli Feng, Han Hu

  • 12 upvotes da comunidade
  • Temas: reinforcement learning, generative models, distribution-wise rewards, reward hacking, mode collapse, sample-wise reward functions

Resumo

Resumo original (em inglês), extraído do paper:

A novel reinforcement learning framework for visual generation uses distribution-wise rewards to improve image diversity and quality while addressing mode collapse and computational efficiency issues.

Onde ler

compartilhar: