AI RESEARCH

Nexus: Same Pretraining Loss, Better Downstream Generalization via Common Minima

arXiv CS.LG

ArXi:2604.09258v1 Announce Type: new