Pim de Haan
@pimdehaan
Followers
1K
Following
503
Statuses
127
Machine learning at CuspAI, materials discovery for carbon capture. @pimdh.bsky.social
Amsterdam
Joined April 2009
@JJitsev @johannbrehmer @TacoCohen Ah I see there is one, interesting. Do you think that would be a good test-bed for this kind of study?
1
0
0
@JJitsev @johannbrehmer @TacoCohen Thanks for you comments. We tried hard to be super transparent about all the limitations of our work in the paper and be careful in our conclusions. Which claims do you find unwarranted?
1
0
1
@JJitsev @johannbrehmer @TacoCohen We found the difference in slope of these curves to not be significantly different (according to leave-one-out testing) in this interval. I don't think it makes much sense to speculate what happens to the right of this interval based on these fits.
1
0
2
@eeevgen @johannbrehmer @TacoCohen If I recall correctly, we were thinking of symmetrizing random non-equi nets as ground truth; or n-body problems in D dims; or solving PDEs. We didn't get to anything specific though
0
0
1
@eeevgen @johannbrehmer @TacoCohen Initially, we planned to study many different groups, representations and datasets, hoping to answer your question. However, we found it difficult to come up with a family of (not very synthetic) datasets with varying groups, and expended all our resources on just one dataset.
1
0
3
RT @wellingmax: Congratulations to @johannbrehmer @pimdehaan @TacoCohen ! Thanks for saving equivariance.
0
3
0
RT @tychovdo: Today at NeurIPS, we’ll be presenting our Noether's Razor paper! 📜✨ 📅 Today Fri, Dec 13 ⏰ 11 a.m. – 2 p.m. PST 📍 East Exhibit…
0
4
0
RT @jonas_spinner: Thrilled to announce that L-GATr is going to NeurIPS 2024! Plus, there is a new preprint with extended experiments and a…
0
6
0
@cottascience @johannbrehmer Thanks! We suspect something like that, but couldn't think of good ways of testing that. We couldn't find realistic/relevant tasks with sufficiently varying symmetry groups.
0
0
2
RT @nicoli_kim: Another exciting day with @msalbergo giving his perspective on generative model for physics, followed by @pimdehaan with ex…
0
4
0
RT @miniapeur: 90% of researchers at big techs: We have multiple internship positions available in my team. Come and work with us on LLMs a…
0
37
0
Coming to NeurIPS: learning the symmetries of dynamical systems with Noether's theorem! Congrats to amazing first author @tychovdo
🌟New work: Noether's razor⭐️ Our NeurIPS 2024 paper connects ML symmetries to conserved quantities through a seminal result in mathematical physics: Noether's theorem. We can learn neural network symmetries from data by learning associated conservation laws. Learn more👇. 1/16🧵
1
10
110