Mila’s annual supervision request process is open for MSc and PhD applications for Fall 2027 admission. While I won’t be taking on new students for supervision this cycle, I encourage you to apply to join Mila’s brilliant deep learning research community. For more information and guidelines, you can visit the link below or contact the @Mila_Quebec team directly. mila.quebec/en/prospective-s…

I recently moved to TU Darmstadt (Germany) and am currently hiring PhD student through ELISS. ellis.eu/news/ellis-phd-prog… If you interested, please apply and choose me as a potential advisor. Topics include Continual/Distributed/Federated/Active/Transparent AI.

Mila brings together over 2,000 researchers in AI and machine learning, an extraordinary concentration of talent in one research institute. Want to be part of this community? Applications for MSc and PhD programs are now open!

Introducing RoboJEPA, the first scaling law for multiembodiment world models. We scaled JEPA predictors up to 8B params, showed scaling laws extrapolate and correlated compute scaling with planning performance. We release all checkpoints and the code. robojepa.github.io

Big new paper nobody is talking about ‼️ 🌎 Robot world models now have scaling laws. Meta trained an 8B-parameter JEPA on 15,000 hours of robot video to prove it. RoboJEPA (FAIR at Meta + Mila) imagines the future in V-JEPA 2.1's feature space, then plans toward a single goal image. The authors call it the largest JEPA predictor trained to date. - Trained on 23 public datasets: 12 robot embodiments, 15,022 hours of video, 6,692 of them with actions. - Fit the scaling law on 22M–2B models and it accurately predicts the 4B and 8B results. - Skills unlock in order as compute grows: moving the arm, holding an object, avoiding obstacles, and only past ~10^22 FLOPs, pushing objects. - On a real Franka with no task fine-tuning, the 8B model grasps 67% of the time vs 5% for π0.5. Not apples-to-apples (goal image vs text), and π0.5 still wins pick-and-place, 53% vs 27%. Why it matters: SCALING LAWS. You can forecast what more compute buys before you spend it.

We (@ChandarLab) are continuing our tradition of helping underrepresented students apply for grad school by reviewing their application materials and giving feedback! Please retweet this for maximum reach! Deadline Nov 1st!

I think once in a while we should stop projecting too much and all just calm down a bit about whether it still makes sense to do research or do a PhD - to my knowledge AI still hasn't cured any major disease yet.

I refuse to play that game :) The joy of science is to pursue your curiosity and answer questions *you* find interesting, not what an instruction-tuned LLM finds interesting, or to play some "game". Might as well go to corporations and get paid more to play "games".

🎉 Thrilled to share that 8 papers from our lab (and collaborations with lab members) have been accepted at #NeurIPS2026! The work spans RL, world models, dynamical systems, LLM safety & interpretability, tabular foundation models, and forecasting. Huge congrats to everyone 🧵👇

re "LLMs vs stochastic parrots" debate, including "but current LLMs are not just LMs anymore" and "but Emily, Timnit etc still make these claims": why do you care so much what a couple of irrelevant scholars say? take any topic, you will find an idiot scholar making bad claims

Our semantic world models paper is accepted at #NeurIPS2026 E&D Track! 🎉 What's the best latent space for diffusion world models? We find semantic spaces beat VAE-style ones on planning and robustness, with competitive visual quality. Paper & more: hskalin.github.io/semantic-w…

Safety and alignment are continual learning problems. AI models must be "raised" with constant feedback that lets them correct their mistakes, just like kids.

An incredible few days at CoLLAs'26 in Bucharest 🇷🇴✨ The audience was very engaged, and the questions after my talk, during the panel and mentorship session made for some fun, deep conversations about where CL is heading. A giant shout-out to the organizers and the local team 🚀

📢 Is Continual Learning trapped in obsolete abstractions? 🤔 We kicked off Day 4 of #CoLLAs2026 with a fantastic keynote from Amos Storkey on "The Art of Forgetting." We moved beyond old benchmarks to see how forgetting impacts modern AI frontiers! 🌍🤖 🎙️ Huge thanks for an amazing talk and insightful discussions!

📢 We’ve just wrapped up Poster Session II at #CoLLAs2026! 🚀 A huge thank you to everyone who presented and attended. We loved the insightful discussions and we hope everyone sparked some productive new connections! 🌍🧠

📢 How do we build agents that know what they know? 🤔 At #CoLLAs2026, Joseph Campbell’s Early Career Talk explored introspection for lifelong agents. We learned that interpretability is more than transparency: it's a core mechanism for agents to reuse & acquire knowledge! 🤖✨ 🎙️ Huge thanks for the amazing talk!

In our second keynote of Day 3 at #CoLLAs2026, Claire Vernade unpacked the online learning foundations of Continual Learning, using dynamic regret & explicit resource constraints to formalize non-stationarity and adaptation. 🧠📊 🎙️ Huge thanks for the inspiring talk!

📢 We continued our Day 3 at #CoLLAs2026 with a wonderful session of Oral presentations, covering: 🧠 NORACL: Neurogenesis for Oracle-free Resource-Adaptive Continual Learning (Karthik Charan Raghunathan, Christian Metzner, Laura Kriener, Melika Payvand) 🎯 TOOD: Task-Aware Out-of-Distribution Score Calibration for Continual Learners (Mostafa ElAraby, Samer B. Nashed, Liam Paull) ⚡ Beyond Single-Model Optimization: Preserving Plasticity in Continual Reinforcement Learning (Lute Lillo, Nick Cheney) 🌐 Federated Continual Learning as a Distributed Drift-Plus-Penalty Control Problem (Nazreen Shah, Naveen Kumar Reddy Somireddy, Zubair Shaban, Ranjitha Prasad, Bharath B N) Congrats to all the authors on their fantastic presentations! 👏🚀

📢 A big thank you to Saba Ahmadi and our Platinum sponsor @Huawei for their insightful presentation at #CoLLAs2026! 🚀 It was great to look at their latest research on attention modulation, retrieval memory control for Long Context Transformers and Trajectory-Balanced Post-Training for Diffusion LMs. Thank you to the @Huawei team for supporting the Lifelong Learning community! 👏