Robin Walters

@RobinSFWalters

Asst. Prof. at Khoury College of CS at Northeastern

Boston, MA robinwalters.com Joined February 2020

209 Following 508 Followers 111 Posts

Latest posts 5

Sometimes the cure can be worse than the disease. Oversmoothing in GNNs is a big issue. Unitarity prevents all smoothing, fixing this. But for many dynamics problems, we want some smoothing! Relaxed unitary convolutions are the sweet spot.

Eddie Berman @bermanatorCodes Jun 29

I'll be @icmlconf presenting our work Smoothness Errors in Dynamics Models and How to Avoid Them, with Luisa Li, @park_jungy , and @RobinSFWalters !!

Come see our poster #2200 , july 9th, 2:30 PM – 4:15 PM, hall A.

Links to paper, website, and code in thread!

View quoted post on X

Robin Walters @RobinSFWalters Jun 30

This work evolves our previous work LieGAN (arxiv.org/abs/2302.00236) which had some optimization issues, especially for discrete groups. LieGAN is great, but it stored only a point estimate of the symmetry generators at any given time.

arxiv.org Generative Adversarial Symmetry Discovery Despite the success of equivariant neural networks in scientific applications, they require knowing the symmetry group a priori. However, it may be difficult to know which symmetry to use as an...

Jung Yeon Park @park_jungy Jun 29

(1/8) Happy to share that our paper LieFlow has been accepted to #ICML2026.

Symmetry is a strong inductive bias in ML, but you have to know the symmetry in advance. Can a model discover it directly from data, for both discrete and continuous groups? 🧵

View quoted post on X

Robin Walters @RobinSFWalters Nov 27, 2025

ISP can do equivariant learning from only RGB images. We leverage the spherical projection technique we previously used in Image2Sphere for prediction.

Boce Hu Verified account @boce_hu Nov 27, 2025

Coauthors: @Dian_Wang_, @_davidklee, @HengT0812, @XupengZ, @HaojieHuang13, @RobotPlatt, @RobinSFWalters.

More details can be found in:
Website: isp-3d.github.io
Paper: arxiv.org/abs/2505.16969
Video: youtu.be/b1aAnbDHQh0?si…
View quoted post on X

Robin Walters @RobinSFWalters Jun 18, 2024

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.

Haojie Huang Verified account @HaojieHuang13 Jun 18, 2024

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/

View quoted post on X

Robin Walters @RobinSFWalters Jun 18, 2024

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

Sharvaree Vadgama Verified account @SharvVadgama Jun 17, 2024

@GRaM_workshop decisions are out now! Woohoo! 🎉🎉 What a delight it was to read your 🌟brilliant papers 🌟!
We are thrilled to announce that we now have some great accepted extended abstracts and wonderful full papers accepted in the proceedings track to be published in PMLR.
View quoted post on X