Reproduce “Beyond Private Training: The New Landscape of AI Privacy” 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
Sean Culatana, Kang Li
The authors audit deletion in retrieval systems by checking which graph nodes search scores, not only which identifiers it returns. Their traversal-safe method skips deleted nodes before distance calculations and repairs live-node connectivity. In Faiss, output filtering leaves distance computations unchanged; at 70% deletion, traversal-safe search improves Recall@10 over Faiss by 16.2 percentage points for targeted deletion, but not random deletion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.