Reproduce “Mitigating Membership Inference Attacks by Self-Distillation Through a Novel Ensemble Architecture” 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
Xinyu Tang, Saeed Mahloujifar, Liwei Song, Virat Shejwalkar, Milad Nasr, Amir Houmansadr, Prateek Mittal
Researchers from Princeton University and the University of Massachusetts Amherst developed SELENA (SELf ENsemble Architecture), a machine learning framework for training privacy-preserving models against membership inference attacks (MIAs). This approach achieves a superior balance between model utility and membership privacy, limiting MIA accuracy to 54.3-58.3% across datasets while incurring a maximum 3.9% drop in classification accuracy without relying on external public data.
Key insights
- Enforcing similar model behavior on member and non-member inputs at inference time can effectively mitigate direct membership inference attacks.
- The privacy-preserving properties of an ensemble can be distilled into a single, compact student model, making the defense robust against adaptive and multi-query attacks while maintaining efficiency.
- Leveraging self-distillation (generating soft labels from the same dataset) eliminates the practical dependency on external public datasets for privacy-preserving knowledge transfer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.