🔥Telegram İfşa

📢 Check out our paper titled "Epidemic Learning: Boosting Decentralized Learning with Randomized Communication" at #NeurIPS2023! 📷 Poster Session: Great Hall & Hall B1+B2 (level 1) #1214 Thu 14 Dec, 19:00 CST. Full paper: openreview.net/forum?id=ytrhs…)

I'm glad that the research of Calicarpa SÀRL's co-founders, El Mahdi El Mhamdi, Sébastien Rouault and myself, was regarded by National Institute of Standards and Technology as a state-of-the-art to mitigate #cyberattacks on #AI training. csrc.nist.gov/pubs/ai/100/2/…

A periodic reminder to reviewers: If you ask authors for more experiments, then you need to communicate a clear hypothesis you're trying to verify with those (e.g., effectiveness on imbalanced data, generalization beyond a certain modality, scalability, etc). Otherwise don't!

C'est finalement allé assez vite ! L'article de recherche, co-écrit Julien Fageot oscar Sadegh Farhadkhani, sera publié à #AAAI2024, et les algorithmes sont déployés sur Tournesol 🌻 depuis septembre 2023. Une science qui fait très plaisir !! arxiv.org/abs/2308.08644

I'm really disappointed with ICLR 2026. Our paper received positive reviews with high scores (8, 8, 6, 5). The AC rejected it based on reasons that are factually incorrect. I've reached out to the PCs about this issue and am still awaiting a response after 2 weeks.

This seems like a clear case of abuse of power from an Area Chair in peer reviewing. ICLR 2026 Program Chairs have a duty to hold Area Chairs accountable to defend scientific integrity. (I am not involved in the paper. My concern is really about scientific integrity in ML...)

A nice thing about doing a phd in Switzerland is that you can have your meetings on the slopes.

The paper is now accepted at #ICML2024. 🥳 Finally, I'll be able to present one of my papers in person. Very happy about that. paper: Tackling Byzantine Clients in Federated Learning. link: arxiv.org/pdf/2402.12780

Introducing "cads: unleashing the diversity of diffusion models through condition-annealed sampling" TL;DR: CADS is an inference method to increase the diversity of diffusion models while maintaining high quality. ETH Zurich AIT Lab #ICLR2024 ait.ethz.ch/cads see the 🧵

I am at #ICML2024. Today I'm presenting our work on Byzantine-Robust Federated Learning, with my amazing colleagues. Come check out our poster from 1:30 to 3 pm, Hall C #1206. paper: arxiv.org/pdf/2402.12780

It has not been reported much, but I believe ETH Zurich has, as of last week, banned new Master and PhD students who attended a long list of universities in China, Russia, and Iran. 🧵

San Diego? or… Brazil!? I’m just saying, my papers might need a bit more polishing.

I believe successful neural network training represents cases of "near convexity": the optimization landscape, while technically non-convex, behaves enough like a convex problem that standard convex optimization is often applicable. At the same time, *in general* neural nets

This is what they call "an agent."

We are hosting a student researcher this year at the Paradigms of Intelligence team at Google! Interested in working with Nino Scherrer and me on AGI, or whatever you think is the next big thing 🥰, please consider applying! docs.google.com/forms/u/2/d/e/…

I went into academia to search for truth. I left learning how to write safe grants, avoid controversy, and serve bureaucrats.

If you are interested in safety/security jailbreaking of LLMs, defenses against them, and how the safety issues become more complicated when we design agentic workflows, this tutorial by Hamed Hassani, Amin Karbasi, Alex Robey is highly recommended

This calculus joke involving potatoes.😂

If it smells fishy, it most likely is a fish. I don't like fish.