←
Tudor Cebere142 posts
Tudor Cebere avatar

Tudor Cebere

@TCebere

PhD 🧑🏻‍🎓 in Differential Privacy & Learning @Inria 🇫🇷 Alumnus @ENSdeLyon 🇫🇷, @upb1818 🇷🇴

◷ Joined July 2020
Sasho Nikolov (thesasho@bsky.social) avatar
⟳ Tudor Cebere retweeted
Sasho Nikolov (thesasho@bsky.social)@thesasho·Aug 4

Nikhil Bansal and I wrote a book-length survey of algorithmic and geometric methods in discrepancy theory, and a draft of it is online arxiv.org/abs/2608.00140. This has been years in the making, and I’m excited to share it. More 👇

Alex Meterez avatar
⟳ Tudor Cebere retweeted
Alex Meterez@alexmeterez·Jul 29

1/8 How well does the simplest model in optimization theory, the quadratic, actually describe LLM pretraining? Surprisingly well. We linearize a 150M model at checkpoints throughout training and the local Taylor expansion tracks the true loss for up to 10% of the training budget.

Bastian Grossenbacher-Rieck avatar
⟳ Tudor Cebere retweeted
Bastian Grossenbacher-Rieck@Pseudomanifold·Jul 29

Our @NeurIPSConf #NeurIPS submission still has *no* meta-review. How cooked are we? Seriously: Can the hivemind help and let us know that someone is looking into this (and that maybe we are not the only ones)? [Please amplify for reach.]

Mohammad Yaghini avatar
⟳ Tudor Cebere retweeted
Mohammad Yaghini@MYaghini·Apr 25

Can ML models be trained under fairness constraints with formal DP guarantees, without sacrificing utility? New paper at ICLR 2026: RaCO-DP Poster: Pavilion 4 (later today!) 🧵 (w/ @tudorcebere, Michael Menart, Aurélien Bellet, @NicolasPapernot )

Ilia Shumailov🦔 avatar
⟳ Tudor Cebere retweeted
Ilia Shumailov🦔@iliaishacked·Jan 19

AI agents face a critical security threat: Indirect Prompt Injection (IPI) attacks hijack agent behavior to steal credentials and cause financial loss. We built the first system-level defense for Computer-Use Agents that actually maintains utility! That is, don't change the model, change the system around it to guarantee control flow integrity! arxiv.org/pdf/2601.09923

Post attachment
Tudor Cebere avatar
Tudor Cebere@TCebere·24 Oct 2025

Grateful to be named a 2025 Google PhD Fellow in privacy, safety, and security! I am excited to continue my research in differential privacy, auditing, and trustworthy AI.🎉🎉

Google.org@Googleorg23 Oct 2025

🎉 We're excited to announce the 2025 Google PhD Fellows! @GoogleOrg is providing over $10 million to support 255 PhD students across 35 countries, fostering the next generation of research talent to strengthen the global scientific landscape. Read more: goo.gle/43wJWw8

Post attachment
Tiffany Ding avatar
⟳ Tudor Cebere retweeted
Tiffany Ding@tifding·10 Jul 2025

New paper: Conformal prediction for long-tailed classification🐒 arxiv.org/abs/2507.06867 🧑‍🌾 (plant enthusiast): Help me identify plants! 🤖 (existing conformal algs): Do you want sets that never include rare plants or sets that contain 100s of labels? 🧑‍🌾: Uhh… neither? A🧵(1/n)

Gautam Kamath avatar
⟳ Tudor Cebere retweeted
Gautam Kamath@thegautamkamath·30 Jun 2025

ICML's election for their board of directors has begun. I've thrown my hat in the ring. Please consider voting for Gautam Kamath. I have experience with the governance of TMLR, COLT, and ALT, and I think I've demonstrated myself as a consciencious and engaged community member.

Post attachment
Alex Meterez avatar
⟳ Tudor Cebere retweeted
Alex Meterez@alexmeterez·21 Jun 2025

Learned a lot by working through this! Check out Sham's tweet summarizing our work and the full note

1/6 Infinite-dim SGD in linear regression is the strawman model for studying scaling laws, critical batch sizes, and LR schedules. We revisit (and simplify) its analysis using just linear algebra, making it easier to derive and reason about. No PSD operators. No tensor calculus.

Roy Rinberg avatar
⟳ Tudor Cebere retweeted
Roy Rinberg@RoyRinberg·9 Jun 2025

RAGs are extremely useful, and yet there isn't an opensource RAG system for wikipedia (or I couldn't find it). So I built WikiRAG, a simple open-source github + hugging repo✨ Spin up your own RAG server for wikipedia in a single line. 🚗💨

Post attachment
OpenDP avatar
⟳ Tudor Cebere retweeted
OpenDP@opendp_org·8 May 2025

SAVE THE DATE for the 2025 OpenDP Community Meeting on September 19th in Dublin, Ireland! We are thrilled to co-locate with the Eyes-Off Data Summit (@Oblivious) and can't wait to share more information on the agenda in the coming weeks - stay tuned 👀 sites.harvard.edu/opendp/eve…