My course on differential privacy is public! Check out the course page for: lecture videos 🎥, lecture notes 📝, and suggested readings 📖! gautamkamath.com/CS860-fa2020.h… Be sure to subscribe on Youtube to catch the latest lectures! youtube.com/playlist?list=… #privacy

Good paper by Naseri, Hayes, and De Cristofaro on robustness for federated learning with differential privacy. Applying DP stops attacks better than other complex robustness mechanisms. Low main accuracy (10-20% drop) might improve with larger datasets. arxiv.org/abs/2009.03561

Dr. @ChrisBishopMSFT and Mila - Institut québécois d'IA's Professor Yoshua Bengio are two of the most prominent scientists in ML. Join them on Dec 1 for a discussion/live Q&A about their thoughts on AI, plus the challenges and frontiers in the algorithms. Register now: aka.ms/AAadfdv

Excited to be selected for this! I'm joining the Shadow PC for #SP21.

The Devil's Guide to Incremental Research In the past years, I have noticed a number of "recipes" for scientific papers – patterns that maximize the number of publications despite limited significance. To raise awareness for these recipes, we have to name them. Enjoy! (1/18)

Introducing a report that summarizes the past 10 years of progress toward building secure AI systems. The report makes this problem more tangible by collecting signals from academia, industry, and governments. #AISecurity #TrustworthyAI #ResponsibleAI adversa.ai/report-secure-…

As I struggled to deal with the impact of COVID on my family members in India, I got delayed by a day for submitting my reviews for a conference & I got a message from a senior reviewer with the blurb below. My humble request to everyone - pls don't say this to anyone ever! [1/n]

Saarland University introduces new bachelor’s degree courses in computer science taught entirely in English – and is the first German state university to do so (Photo: O. Dietze) Saarland Informatics Campus : idw-online.de/de/news767840

Mathematics for Machine Learning Nice book. PDFs of this book are freely available. mml-book.github.io Syntest two, Analog Aldo, @[email protected] on Mastodon #MachineLearning #datascience #mathematics

"Since few years, the quality I now appreciate most in papers is honesty: Here’s what we did, here’s what it does, here’s when we know it works, here’s what will not work." This is a great article by Andreas.

I'm excited to join Microsoft Research Confidential Computing Lab this coming week as a research intern! Looking forward to working with a fantastic team.

Join us this Thursday at 1pm BST (8am EDT/8pm CST) at the workshop on Theoretical and Practical Privacy of ML Training Pipelines. We have a fantastic lineup of invited speakers and we will end with an open discussion. Registration is open and free for all: researchsummit.microsoft.com

Congrats to Mohammad Naseri and @_jamiedh for getting their paper on privacy in FL accepted at NDSS Symposium !

Help needed! We are searching industrial practitioners or ML-researchers for our survey about Adversarial ML in practice. Participation is anonymous and just takes 10 minutes, but your opinion is highly valuable to us! 👉 Go to survey: forms.gle/iGbZXHgzY1yb7a… Thanks!

If you are a faculty reviewing Iranian applicants, please ignore this guide! This is totally wrong and misleading!

My first in-person conference since I started my Ph.D. #NDSS2022

This week, I will present (remotely) our paper "CERBERUS: Exploring Federated Prediction of Security Events" ACM CCS 2025. Please reach out if you want to have a chat.

I'll be presenting our work "Bayesian Estimation of Differential Privacy" at #ICML2023. Join me from 1:30pm at the poster session (poster #708) to learn how to estimate the actual privacy of ML training pipelines using membership inference attacks.