Reproduce “Difficulty Calibration Across Time for Membership Inference in Diffusion Models” 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
Michael Joseph, Yedid Hoshen
The authors introduce Temporal Self-Calibration (TSC), which compares an image’s denoising loss at one noise level with losses at two symmetric noise levels to reduce image-specific difficulty. On CIFAR-10, TSC detects 62.31% of members at a 1% false-positive rate, versus 30.90% for PIAN, using 15 queries rather than two. Results vary across datasets, and the attack requires intermediate denoising outputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.