Keeping you informed of the latest research advances in AI/ML for Science and Simulations
Germany
Joined December 2022
- Tweets97
- Following34
- Followers898
- Likes124
AL4PDE codebase is available @ github.com/dmusekamp/al4pde. Try it out!
AI4Science Talks retweeted
🚨ICLR poster in 1.5 hours, presented by Daniel Musekamp:
Can active learning help to generate better datasets for neural PDE solvers?
We introduce a new benchmark to find out!
Featuring 6 PDEs, 6 AL methods, 3 architectures and many ablations - transferability, speed, etc.!
AI4Science Talks retweeted
If you are at ICLR and interested in ways to make denoising diffusion more efficient, please come to Vinh’s oral talk tomorrow at 11:30 in oral session 1C.
It also involves backprop through ODE solvers and constrained learning.
🚀 Exciting news! Our paper "Learning to Discretize Diffusion ODEs" has been accepted as an Oral at #ICLR2025! 🎉
[1/n]
We propose LD3, a lightweight framework that learns the optimal time discretization for sampling from pre-trained Diffusion Probabilistic Models (DPMs).
AI4Science Talks retweeted
📰 blogpost:
maxxxzdn.github.io/blog/cscn…
🕹️ google colab:
colab.research.google.com/dr…
I tried to make the blog post more accessible than the paper and added a lot of supporting visualizations.
Please check it out if you are curious about spacetime-equivariant CNNs 🚀
Excited to introduce Clifford-Steerable CNNs: a framework that expands equivariant CNNs to pseudo-Euclidean groups, including the Poincaré group - the group of isometries of spacetime!
Joint work w/ @djjruhe, @maurice_weiler, @__alucic, @jo_brandstetter, and Patrick Forré
1/12
📢 Active Learning for Neural PDE Solvers (AL4PDE; arxiv.org/abs/2408.01536)
While active learning is common in other domains, it has yet to be studied extensively for neural PDE solving. This work introduces AL4PDE, a modular and extensible active learning benchmark.
#ML4PDE
ALT Workflow diagram of active learning for neural PDE solving
Shows that active learning reduces the average error by up to 71% compared to random sampling & significantly reduces worst-case errors.
Acquired datasets are reusable, providing benefits for surrogate models not involved in the data generation.
Code and dataset available soon.
A new line of research towards developing algorithms & methods that enable neural operators to achieve generalization over desiderata such as PDE parameters, ICs, BCs, Geometries, Resolutions, etc.
#NeuralOperators #ML4Science #PDEs #NumericalSolvers #ActiveLearning #AI4Science
AI4Science Talks retweeted
Neural Spectral Methods (spectral-based neural operator + spectral loss training method) will be presented at Poster session 4 this Wed (May 8, 4:30 PM) at #ICLR2024!
Paper: openreview.net/forum?id=2DbV…
Code: github.com/ASK-Berkeley/Neur…
1/ Interested in ML + PDEs? Introducing Neural Spectral Methods, a neural network framework for solving parametric PDEs, inspired by the advantages of classical spectral methods.
Paper: arxiv.org/abs/2312.05225
Hybrid Models for Neural PDE Solving & SciML Simulations 🔥 by @ask1729 #ICML2024 #ML4Science #ML4PDEs
#ML4Science and #ML4Simulations related workshops @ #ICML2024 today:
Machine Learning for Life and Material Science: From Theory to Industry applications (ml4lms.bio/)