Reproduce “Probing Local Conditional Fit for Diffusion Model Membership Inference” 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
ByungHee Cha, Naoki Murata, Yuhta Takida, Jong Chul Ye, Yuki Mitsufuji
LoFIT tests whether an image-caption pair was included in a diffusion model’s training data by measuring how denoising loss initially responds when the text condition changes at fixed noisy inputs and target noise. It averages prediction-direction responses across reference captions and states, requiring denoiser queries. On fine-tuned SD v1.4/Flickr30k, AUROC is 0.935 versus 0.906 for normalized loss gap; generation-only access is insufficient.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.