🚀Proud to share our work on the training dynamics in Transformers with Wassim (Wes) Bouaziz @ ICLR 2025 & @CabannesVivien Inria AI at Meta! 📝Easing Optimization Paths arxiv.org/pdf/2501.02362 (accepted @ICASSP 2025 🥳) 📝Clustering Heads 🔥arxiv.org/pdf/2410.24050 🖥️ github.com/facebookresear… 1/🧵

🚀 Policy gradient methods like DeepSeek’s GRPO are great for finetuning LLMs via RLHF. But what happens when we swap autoregressive generation for discrete diffusion, a rising architecture promising faster & more controllable LLMs? Introducing Score Entropy Policy Optimization

Come work with us on tslearn in beautiful Rennes! (deadline for application is soon!) jobs.inria.fr/public/classic…

Living organisms organize as loosely coupled hierarchies—while our AI systems remain rigid monoliths. What if we built AI that scales and adapts like biological systems instead? What would an AI like this look like? 🧵 1/9

While President Trump was speaking to Congress about Russia’s signals for peace, Russia attacked Ukrainian cities with over 180 drones and ballistic missiles, destroying homes and killing civilians. The only signals Ukrainians heard tonight were air raid sirens and explosions.

Countless emails asking how Americans can directly aid Ukrainian soldiers. Here’s an easy answer: savelife.in.ua/en/

The death toll won’t wait. While the Trump administration has paused U.S. military aid to Ukraine, and medical supplies were withheld from military logistics, frontline medics are still fighting to save lives. Support Leleka Foundation to help them. givebutter.com/AYZpPQ

This is incredible!! Way to go America 🇺🇲 Supporters of Ukraine have unfurled the world's largest Ukrainian flag on the White House ellipse, pushing for the U.S. to continue its aid against the russian invasion: "Do not abandon Ukraine!"

Powerful speech by a French senator. youtube.com/shorts/TEjKctH…

Full speech subtitled in English. youtube.com/watch?v=e-s4BI…

Our ML Theory group is looking forward to welcoming Ambroise Odonnat next week on Thursday, June 19th for a session on "Large Language Models as Markov Chains"

How can we circumvent data scarcity in the time series domain? We propose to leverage pretrained ViTs (e.g., CLIP, DINOv2) for time series classification and outperform time series foundation models (TSFMs). 📄 Preprint: arxiv.org/abs/2506.08641 💻 Code: github.com/ExplainableML/…

Excited to announce a call for papers for our workshop on time series foundation models NeurIPS Conference, with an amazing lineup of speakers. See the post below for more details on where to submit (deadline: 22nd of August).

Following numerous requests, we've decided to extend the submission deadline for the NeurIPS Conference workshop on Recent Advances in Time Series Foundation Models! 📜New deadline: August 29 🌐 Openreview link: lnkd.in/ed2T3vaR