WARP: Weight-Space Analysis for Recovering Training Data Portfolios

WARP: Weight-Space Analysis for Recovering Training Data Portfolios

WARP is a framework that infers training data compositions from released model weights by analyzing geometric footprints in weight space through model merging and feature extractio…

Hugging Face · Daily Papers ·Tzu-Heng Huang, Aditya Goyal · ·▲ 2 upvotes

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

Autores: Tzu-Heng Huang, Aditya Goyal, John Cooper, Frederic Sala

  • 2 upvotes da comunidade
  • Temas: foundation models, model merging, pseudo-checkpoints, weight space, geometric footprint, geometric features

Resumo

Resumo original (em inglês), extraído do paper:

WARP is a framework that infers training data compositions from released model weights by analyzing geometric footprints in weight space through model merging and feature extraction.

Onde ler

compartilhar: