A really nice paper on lower bounds for private estimation by Gautam Kamath ✈️ NeurIPS 2025, Argyris Mouzakis, and @vkerdos. 1/4 arxiv.org/abs/2205.08532

ICML 2022 Accepted Papers: icml.cc/Conferences/20… #ICML2022 ICML Conference

3 papers on efficient and differentially private learning of Gaussians in #colt2022 ! Interestingly each paper uses a different approach. Core challenge is handling covariances with high condition numbers. You cannot simply add isotropic noise; it has to "scale" with data [1/5]

#NeurIPS2022 Accepted Papers: nips.cc/Conferences/20… NeurIPS Conference

⚠️🚨BONUS TWEET 🚨⚠️ Be sure to also check out this other paper posted at the same time, by Sasho Nikolov ([email protected]) and Haohua Tang, also focused on unbiased algorithms in differential privacy. Despite similarities in the titles, the settings are mostly different. arxiv.org/abs/2301.13850 9/8

Special congratulations to Matt (currently a Master's student - he's not on twitter afaik) who, out of the junior people, is the person who lead this project!

Congrats to Waterloo's Cheriton School of Computer Science colleagues Niki Hasrati (Niki Hasrati) & Shai Ben-David (shai ben-david) on best paper at #ALT2023. Main result shows when an online learner must be computable, it's no longer characterized by Littlestone dimension. Check it out: arxiv.org/abs/2302.04357

I was at a sensational combinatorics seminar in Cambridge yesterday, reminiscent of the time I had been tipped off that Andrew Wiles's seminar at the Newton Institute on Wednesday 23rd June 1993 might be worth going to. 🧵 arxiv.org/abs/2303.09521

Congrats to Duchi-Haque-Kuditipudi (arxiv.org/abs/2301.07078) and Brown-Hopkins-Smith (arxiv.org/abs/2301.12250) on #COLT2023 best student paper for their works on differentially private mean estimation! Wanna know what their works are about? Read on 🧵1/n

Thanks Gautam Kamath both for posting this and for advising and supporting me! It's a privilege to be part of The Salon ! Also, thanks to all of my co-authors and mentors for their support (Matt Regehr, @vkerdos, Thomas Steinke and Jonathan Ullman)!

🧵Highlighting a SODA 2024 accepted paper, "Sorting and Selection in Rounds with Adversarial Comparisons," authored *solely* by Waterloo Mathematics undergraduate Chris Trevisan. This is the only single-authored paper by an undergrad in this SODA, a huge achievement!! Read on 👇1/n

🧵New paper: "Not All Learnable Distribution Classes are Privately Learnable" to appear in #ALT2024. We refute a conjecture of Hassan Ashtiani. We show that there exists a learnable distribution class which is not privately learnable. w Mark Bun Argyris Mouzakis @vkerdos 1/n

The simplest possible optimal PAC learner? New arXiv preprint: arxiv.org/abs/2403.08831

After a long but very thoughtful review and revision process, this paper was finally accepted to the Journal of the American Statistical Association (JASA)! Congrats to the junior authors Matthew Regehr (Matt Regehr), Argyris Mouzakis (Argyris Mouzakis), and Vikrant Singhal (@vkerdos).

🧵New paper on arXiv: Optimal Differentially Private Sampling of Unbounded Gaussians. With Waterloo's Cheriton School of Computer Science undergrad Valentio Iverson (Valentio Iverson) and PhD student Argyris Mouzakis (Argyris Mouzakis). The first O(d) algorithm for privately sampling arbitrary Gaussians! 1/n

Congrats to my undergrad Valentio Iverson (Valentio Iverson) on winning the Germain-Erdős Undergraduate Award in Mathematical Research, for “remarkable contributions to basic mathematical research." His work includes a COLT 2025 paper on private sampling with me and Argyris Mouzakis 1/3