📢#GRaM template for proceedings track is up on our website gram-workshop.github.io . Submit your great ideas in a #ICML like 8-page submission. Accepted papers will be published in PMLR ✨#GRaM proceedings ✨.

A lot of work on symmetry in deep learning is focused on symmetry in data (e.g. equivariant neural networks), but there's lots of symmetry in parameter space too! You can use this structure to improve optimization algorithm speed and even generalization.

#ICLR24 We proposed FourTran, a very sample-efficient 3D manipulation pick-place model. 1. It can learn a nontrivial 3D policy with less than 10 demos. 2. It represents 3D action distribution in Fourier Space. Check it in the Poster Session 4 at 4:30 PM Vienna time (10:30 EDT)

📣 This year at #ICML2024 we are hosting ✨ Gram Workshop ✨ Geometry-grounded representation learning and generative modeling. We welcome submissions in multiple tracks i.e. 📄 Proceedings, 🆕 extended abstract, 📝Blogpost/tutorial track as well as🏆 TDA challenge.

Convolutional neural nets going to spacetime 🚀 Our new ICML24 paper shows how to build Lorentz-equivariant CNNs/MPNNs for multivector fields on Minkowski spaces. This is useful for particle physics or Navier Stokes / electrodynamics simulations. arxiv.org/abs/2402.14730 🧵1/N

The reviewing quality was exceptional. Thank you to our fantastic reviewers!

Excited to publish our recent work - Imagination Policy: Using Generative Point Cloud Models for Learning Manipulation Policies! It generated the imagination of the key-frame actions and achieved very high sample efficiency. Project Website: haojhuang.github.io/imagine_page/

It's interesting that the much higher dimensional problem of point cloud generation leads to be better performance in the low dimensional problem of finding the best SE(3) pose. It might be point clouds have more geometric structure or another victory for overparameterization.

Introducing Equivariant Diffusion Policy, a novel sample efficient BC algorithm based on equivariant diffusion. Our method leverages the symmetry in policy denoising to boost learning — needing 5x less training data in sim and mastering complex tasks in real-world with <60 demos.

Join us on Monday October. 14th at 2pm (UTC+4) in #IROS2024 Workshop on Equivariant Robotics. A great lineup of keynote speakers will discuss how symmetry penetrates each and every subfield of robotics. Website and Zoom Link: equirob2024.github.io

Grasp detection is crucial for robotic manipulation but remains challenging in SE(3). We introduce our #CoRL2024 paper: OrbitGrasp, an SE(3)-equivariant grasp learning framework using spherical harmonics for 6-DoF grasp detection. 🌐 orbitgrasp.github.io

#CoRL2024 IMAGINATION POLICY: Using Generative Point Cloud Models for Learning Manipulation Policies Led by Haojie Huang. A key-frame multi-task policy can generate key poses (imagine) and do manipulation precisely with sample efficiency. Presenting at Poster Session 4.

#CoRL2024 Looking forward to presenting Equivariant Diffusion Policy at Oral Session 1 on Nov 6 and Poster Session 2 on Nov 7! Also checkout equidiff.github.io for the paper and code.

Generate the goal state and then infer the manipulation pick-place action. Feel free to check poster session 4 at #36 for details.

Honored to receive my first best paper nomination from Conference on Robot Learning! Had such a great time at #CoRL2024, huge thanks to the organizing committee and all my co-authors: Steve, David, Tarik Kelestemur, Haojie Huang, Haibo Zhao, Mark, JW, Robin Walters, Robert Platt!

What can we learn from neural network model weights? Join us for the Weight Space Learning Workshop at #ICLR2025! ICLR 2026 📄Accepting extended abstracts & full papers 🗓️Submission Deadline: Feb 4, 2025 🔗weight-space-learning.github.io

Save the date -- Boston Symmetry Day 2025 will be held on March 31st, at Northeastern University! Speakers and sponsors to be announced in the coming weeks, but you can expect another great lineup of talks, networking, and posters. We'll see you there!

Registration is now open for Boston Symmetry Day on March 31! Sign up by March 21st at docs.google.com/forms/d/e/1FAI… We have an exciting lineup of speakers (see our website: bostonsymmetry.github.io )  Also featuring a poster session so you have a chance to present your awesome work!

When and why are neural network solutions connected by low-loss paths? In our #ICML2025 paper, we show that mode connectivity often arises from symmetries—transformations of parameters that leave the network’s output unchanged. Paper: arxiv.org/abs/2505.23681 (1/6)

Excited to share our #ICML2025 paper, Hierarchical Equivariant Policy via Frame Transfer. Our Frame Transfer interface imposes high-level decision as a coordinate frame change in the low-level, boosting sim performance by 20%+ and enabling complex manipulation with 30 demos.