Reproduce “On the Relationship between Model Quantization and Model Inversion Attacks” 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
Rongke Liu, Youwen Zhu
The method refines post-training quantization without retraining network weights, using task-sensitive bit allocation and geometry-regularized scale and rounding updates to recover recognition utility. On Palm with ResNet-50 and four-bit weights and activations, refined accuracy is 96.55%, versus 99.01% for FP32, while RL-MIA Strict success falls from 54% to 26%. These results cover tested models and attacks, not general privacy protection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.