Reproduce “Revisiting Model Stitching In the Foundation Model Era” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zheda Mai, Ke Zhang, Fu-En Wang, Zixiao Ken Wang, Albert Y. C. Chen, Lu Xia, Min Sun, Wei-Lun Chao and 1 more
The authors introduce model stitching, connecting early layers of one vision model to later layers of another through a trainable bridge, and test it across differently trained models. They train the bridge to match the target’s final features before task adaptation. In tested settings, this beats self-stitch controls; an early VST evaluation recovers 45.5% of two-model gain at 4.3% added resource cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.