Reproduce “Weight Space Representation Learning via Neural Field Adaptation” 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
Zhuoqian Yang, Mathieu Salzmann, Sabine Süsstrunk
The authors introduce a shared-base method for representing images and shapes with neural-field weights. It freezes a pretrained field and fits each instance using low-rank multiplicative adaptation, then trains diffusion models over the adapted weights. On 128 × 128 FFHQ faces, mLoRA-Asym achieves a Fréchet distance of 0.073 versus HyperDiffusion’s 0.241, indicating improved distributional similarity, not state-of-the-art image generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.