Controlling Action-Head Co-Adaptation to VLM Representation Anisotropy
arXiv:2611.04006v1 Announce Type: new Abstract: Modern learning systems increasingly combine pretrained models with downstream heads. Despite the growing importance of such modular designs, how the downstream model co-adapts to the geometry of pretrained representations remains relatively underexplo
Der Artikeltext wird beim nächsten Sync nachgeladen. Original bei arXiv cs.AI lesen
Quelle: arXiv cs.AI (arXiv cs.AI). Original öffnen