Poster session at 3PM today AISTATS Conference :)

SSL is often motivated by the medical domain, where annotated data can be rare.

Théo Moutakanni provides evidence of improved performance, robustness, and generalization in RayDINO.
We now have a wonderful paper to reference to justify this intuition !

arxiv.org/abs/2405.01469

Meta announces Better & Faster Large Language Models via Multi-token Prediction

Large language models such as GPT and Llama are trained with a next-token prediction loss. In this work, we suggest that training language models to predict multiple future tokens at

Tom Sander (Tom Sander) from CMAP (@Polytechnique Institut Polytechnique de Paris ) and meta (@MetaFrance) led the team meeting on Monday! Thanks for your interesting presentation on
'Watermarking Makes Large Language Models Radioactive' !

Giovanni Neglia and I are hiring a PhD student in #FederatedLearning ! Details here: jobs.inria.fr/public/classic… The precise topic is flexible and can be adapted depending the skills and interests of the applicant. Please RT!

🚀🚀Check out the final version of our spotlight paper at ICLR 2024 on the convergence of the hidden states of Residual Networks to the solution of a Neural ODE! 🚀🚀

Paper: arxiv.org/abs/2309.01213

Code: github.com/michaelsdr/imp…

with Pierre Marion, Yuhan Wu and Gérard Biau

Yes it does ☢️☢️☢️
Check out our paper: arxiv.org/abs/2402.14904

Meta presents Watermarking Makes Language Models Radioactive

paper investigates the radioactivity of LLM-generated texts, i.e. whether it is possible to detect that such input was used as training data. Conventional methods like membership inference can carry out this detection…