Covariance-Aware Private Mean Estimation, Efficiently differentialprivacy.org/colt23-bsp/

Today the 25 National AI Research Institutes (funded by the National Science Foundation U.S. National Science Foundation ) are showcasing in the US Senate. Excited to be part of this event, presenting our work on generative AI in our IFML institute.

Here are a few things I learned from the AI Institutes #AIHillDay showcasing in the Senate yesterday: 0. Many on the Hill are talking about AI. Things are happening. 1. It is not obvious to many in DC that universities play a key role in developing the *research* used in modern

Wow! I honestly don't know what to say. Receiving the NeurIPS outstanding paper award is an amazing honor for me and my coauthors, Milad & Matthew, but also all of the prior research that we built on.

Thanks for attending Desh Raj! Glad to host friends from 3 continents, 4 countries, 5 religions and 3 nationalities 🤗

New! A rich & thought-provoking article by Milan Shen Weijie Su & Andreas Terzis that provides a comprehensive overview of challenges in deploying #DifferentialPrivacy, of building effective DP infrastructure, & of fruitful research directions. doi.org/10.1162/99608f…

Congratulations to Germain and Pulkit Pulkit Tandon! (honorable mention ICLR 2025)

The 2nd Workshop on Generative AI and Law (GenLaw) is going to be at ICML: genlaw.org/2024-icml/cfp.…

Exciting talks and posters in two days at 'Privacy Regulation and Protection in Machine Learning' workshop. pml-workshop.github.io/iclr24/

Happening now in Schubert 3!! Join us for the privacy regulation and protection workshop!

Of course, if your work is on zeroth order optimization, consider CC BY-NC-ND 4.0. After all, there are "No derivatives."

Imagine asking an LLM assistant to book a restaurant. It needs to know your name, phone number, and a credit card number. But does it know not to share your age or SSN? We show that an adversarial third-party can easily extract personal information by hijacking the context. 1/5

I am really excited to introduce DataComp for Language Models (DCLM), our new testbed for controlled dataset experiments aimed at improving language models. 1/x

"Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining," with Florian Tramèr & Nicholas Carlini got an #ICML2024 best paper award! x.com/thegautamkamat… 🧵: the personal side of this research, emotional high & lows, & more 👇 1/n

We have been awarded two Best Paper Awards at ICML Conference 2024 for 'Stealing Part of a Production Language Model' and 'Considerations for Differentially Private Learning with Large-Scale Public Pretraining'! arxiv.org/abs/2403.06634 arxiv.org/abs/2212.06470

Congratulations to Jonathan Hayase on winning the 2024 ICML Best Paper Award for his work during his Google internship.

A fantastic Economist article on Open-source AI by two pioneers: Martin Cassado and Ion Stoica. (plus a nice, human-generated drawing). The main points I got from their Economist article: 1. "Regulation hurts innovation": I agree, and I am worried about Europe going in the

Data poisoning attack is turned into an authentication tool for model ownership in OML.

Playing with Alex Diamkis' new factuality checker from Bespoke Labs: playground.bespokelabs.ai

Congrats Giulia Fanti !