🚨 Preprint Alert 🚀 📄 seq-JEPA: Autoregressive Predictive Learning of Invariant-Equivariant World Models https://t.co/vJaFyoQZvV Can we simultaneously learn both transformation-invariant and transformation-equivariant representations with self-supervised learning (SSL)?

CDS Sr. Research Scientist @cosmo_shirley and CDS Prof. @ylecun are keynoting the 2026 World Modeling Workshop (Feb 4–6 at Mila), with @NYU_Courant's @sherryyangML, and others. The workshop spans SSL, RL, robotics, LLMs & more. 📍 Montréal + online 🔗 https://t.co/94DOjVhgFa

📢Update for the World Modeling Workshop 2026! Unfortunately, Ricky Chen will not be able to give the tutorial on Diffusion/Flow Matching. However, we’re excited to announce that @GlenBerseth has kindly agreed to give a tutorial on VLA instead! 🙌 🌐 https://t.co/inI2YV3bhT

🚨 Interested in generative world models? We’re thrilled to host Stephen Spenser (@GoogleDeepMind) at the World Modeling Workshop 2026, where he’ll talk about the Genie series of models! 🌐 https://t.co/inI2YV2Dsl

🚨 We’re honored to host Prof. Jürgen Schmidhuber (The Swiss AI Lab & KAUST) at the World Modeling Workshop 2026! ✨ A pioneer of modern AI, Prof. @SchmidhuberAI has made influential contributions that shaped the field — we’re thrilled to welcome him. 🌐 https://t.co/inI2YV2Dsl

Excited to share that seq-JEPA has been accepted to NeurIPS 2025!

🌍 Excited to host Florian Bordes (@AIatMeta) at our World Model Workshop! He will present IntPhys 2, a benchmark for AI’s intuitive physics understanding. 🌐 https://t.co/WukFtNOfdQ 📍 Mila, Montreal, Canada 👉 https://t.co/tTC1eAFsL4 #AI #WorldModels #IntuitivePhysics

Super excited to be speaking alongside giants such as @ylecun and @Yoshua_Bengio at the world model workshop 🚀 at Mila! Hope to see many of you in the wondrous Montreal!

We are organizing a hands-on world modeling workshop to discuss, share ideas, and argue about the next generation of world models! An impressive lineup including @ylecun @Yoshua_Bengio @sherryyangML @cosmo_shirley and many more! Submit your work, register, and see you in Feb!

We are organizing a workshop on World Models at Mila next Feb! If you want to understand how to learning and us world models come join the discussion.

Join us to push the field forward! Please spread the word📣📣

Excited to announce this awesome workshop we are organizing at Mila World models touch so many fields from robotics/videos to LLMs and AI4Science: there will be something for everyone! We already have several amazing speakers confirmed, with more to be announced soon 🤩

🚨 New World Models Workshop! Thrilled to be hosting this event @Mila_Quebec! I’m convinced this is exactly what the field needs right now to push our community forward. Please RT + apply if you’re into world models! 🚀

Excited to be one of the organizers of this workshop! If you work on world modeling, join us to discuss the future of the field. Stay tuned for more speakers!

🚨Announcing the World Modeling Workshop 2026 🚨 📅 When: Feb 4–6, 2026 📍Where: Mila (Montréal) + Online (free) 💡 What: Keynotes, Methods Deep Dive, and Tutorials 🌐 https://t.co/WukFtNON3o ✉️ worldmodel.mila@gmail.com 🧵 Details below:

Huge thanks to my supervisors and co-authors @NeuralEnsemble and @ShahabBakht! Check out the full paper here: 📄 https://t.co/vJaFyoQZvV 💻 Code coming soon! 📬 DM me if you’d like to chat or discuss the paper! (7/7)

Interestingly, seq-JEPA shows path integration capabilities – an important research problem in neuroscience. By observing a sequence of views and their corresponding actions, it can integrate the path connecting the initial view to the final view. (6/7)

On 3D Invariant-Equivariant Benchmark (3DIEBench) where each object view has a different rotation angle, seq-JEPA achieves top performance on both invariance-related object categorization and equivariance-related rotation prediction without sacrificing one for the other, as

Seq-JEPA learns invariant-equivariant representations for a variety of tasks that contain sequential observations and transformations; for example, it can learn semantic image representations by seeing a sequence of small image patches across simulated eye movements with no

🧠Humans learn to recognize new objects by moving around them, manipulating them, and probing them via eye movements. Different views of a novel object are generated through actions (manipulations, eye movements, etc.) that are then integrated to form new concepts in the brain.

Current SSL methods often face a trade-off: optimizing for transformational invariance in representational space (useful in high-level tasks, such as classification) often reduces equivariance (needed for fine-grained downstream tasks related to details like object rotation,