@salvaRC7 profile picture
Salva Rühling Cachay @salvaRC7

PhD student @ucsd_cse. Into AI for Science, especially climate&weather.

Joined May 2021
  • Tweets 128
  • Following 804
  • Followers 545
24 Photos and videos
Pinned Tweet
Salva Rühling Cachay
@salvaRC7
8 Dec 2023
Sharing a blog post about our #Neurips2023 paper "DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting" salvarc.github.io/blog/2023/… It's faster than standard diffusion models, has low memory needs, and generates stable & probabilistic rollout forecasts. 1/7
4
14
106
17,850
Salva Rühling Cachay retweeted
Salva Rühling Cachay
@salvaRC7
Jun 4
🌎⚡ A frontier (1.5°, 15-day) ensemble weather forecast in ~3 seconds, from a probabilistic model you can train in ~1 day on a single H200 GPU?* Meet U-Cast, our new #ICML2026 paper. 🧵
1
5
12
2,085
Salva Rühling Cachay retweeted
Jiahe (Chloe) Huang
@jhhuang_chloe
Jun 5
[🚨ICML 2026 Paper Alert] Excited to introduce SimulCost! 🚀 The first cost-aware benchmark and toolkit for evaluating LLM agents on complex physics simulations. Can LLMs efficiently tune complex physics simulators under realistic computational budgets? Spoiler: They struggle! 📉 🔗 Project: sichenglai.com/SimulCost-Web… 📄 Paper: arxiv.org/abs/2603.20253 💻 Code: github.com/Rose-STL-Lab/Simu… 🤗 Models: huggingface.co/datasets/Rose… Huge thanks to the amazing team: @YadiCao, Sicheng Lai, Yang Zhang, Zach Lawrence, Rohan Bhakta, Izzy F. Thomas, Mingyun Cao, Chung-Hao Tsai, Zihao Zhou, Yidong Zhao, Hao Liu, Alessandro Marinoni, Alexey Arefiev, @yuqirose A quick thread 🧵👇
1
2
14
4,454
🌎⚡ A frontier (1.5°, 15-day) ensemble weather forecast in ~3 seconds, from a probabilistic model you can train in ~1 day on a single H200 GPU?* Meet U-Cast, our new #ICML2026 paper. 🧵
1
5
12
2,085
more replies
*Fine-tuning takes ~1 day (1.2 H200-GPU-days) on a pre-trained deterministic backbone; add ~7 H200-days to train that backbone from scratch (~8 total). 📄 arxiv.org/abs/2604.09041 💻 github.com/Rose-STL-Lab/u-ca…

U-Cast: A Surprisingly Simple and Efficient Frontier Probabilistic...

AI-based weather forecasting now rivals traditional physics-based ensembles, but state-of-the-art (SOTA) models rely on specialized architectures and massive computational budgets, creating a high...

arxiv.org
1
75
Together with Duncan Watson-Parris and @yuqirose!
68
Really appreciate the computational support from @modal !!
Charles 🎉 Frye
@charles_irl
Apr 27
some people use @modal to predict the next token, others to predict the next rainstorm arxiv.org/abs/2604.09041
8
956
Salva Rühling Cachay retweeted
Charles 🎉 Frye
@charles_irl
Apr 27
some people use @modal to predict the next token, others to predict the next rainstorm arxiv.org/abs/2604.09041
11
68
4,703
NeurIPS poster session tomorrow morning!⚡️ Come say hi if you are interested in diffusion/AI4Science/PDEs/weather/etc. ⏰ Friday, 11am-2pm PST at Poster #3614 Happy to chat outside these hours as well—I'm here for the full week (and the next ;)
7
21
2,613
🌍 Modeling chaos isn't just about predicting the next step—it's about modeling how uncertainty grows over time.🌪️ I’m thrilled to share Elucidated Rolling Diffusion Models (ERDM), accepted to #NeurIPS2025! We unify rolling diffusion with EDM for forecasting complex systems🧵👇
0:11
1
3
12
1,145
more replies
𝗧𝗵𝗲 𝗥𝗲𝘀𝘂𝗹𝘁𝘀: 3) 𝗣𝗵𝘆𝘀𝗶𝗰𝗮𝗹 𝗥𝗲𝗮𝗹𝗶𝘀𝗺: ERDM matches the power spectra of operational physics-based models (IFS ENS), solving the "blurriness" problem common in AI weather models.
1
116
A huge thank you to my brilliant collaborators @NVIDIAAI -Miika Aittala, @karsten_kreis, Noah Brenowitz, @ArashVahdat & @MardaniMorteza-and @yuqirose @ucsd_cse 👇 See you next week in San Diego! Paper: arxiv.org/abs/2506.20024

Elucidated Rolling Diffusion Models for Probabilistic Forecasting...

Diffusion models are a powerful tool for probabilistic forecasting, yet most applications in high-dimensional complex systems predict future states individually. This approach struggles to model...

arxiv.org
3
267
Load more