Submitted by Siran Peng 320 VisionHOPE: Visual Backbones as Self-Modifying Learning Systems Mininglamp Technology 508 2