Icml_2025

(Belatedly, and happily) sharing that we have two forthcoming papers at ICML 2025 (say hi in Vancouver!),

• the first, on a new (and I think well motivated) threat model for LLMs, Fragments to Facts: Partial-Information Fragment Inference from LLMs is with Bin Han, Robert Wolfe, and Bill Howe.
• and the second, which we had already released as a pre-print, Differential Privacy Under Class Imbalance: Methods and Empirical Insights with Rachel Cummings and Marco Avella Medina.