Reproduce “Deep Model Reassembly” 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
Xingyi Yang, Daquan Zhou, Songhua Liu, Jingwen Ye, Xinchao Wang
DeRy assembles customized networks from blocks of heterogeneous pretrained models, using representation similarity to group compatible blocks and a training-free proxy to rank candidate assemblies under parameter and compute constraints. On ImageNet, the method achieves 78.6% top-1 accuracy with pretrained blocks frozen, while training the stitching layer and classifier; this demonstrates performance in the reported setup, not universal compatibility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.