RT @NYUDataScience: CDS Research Scientist @cosmo_shirley, with @oharub & @mikemccabe210, led the creation of "The Well"—a 15TB dataset for….

Heading off to #NeurIPS2024 today! Feel free to reach out if you're around to catch up! . If you're interested, please visit our poster sessions tomorrow at 11 PST for MPP ( and on Thursday at 11 for the Well (.

This was a huge collaborative effort with too many contributors to list in one post, but it wouldn't have been possible without all of our teammates @PolymathicAI and elsewhere. We hope that access to more realistic problems pushes the development of #AI4Science forward!.

and many more! All data in the Well is stored in a uniform format and whether its 2D or 3D it's accessible through the same API. The Well additionally includes benchmarking tools such as baseline models, tensor-aware augmentations, and metric implementations:

To complex newly understood non-Newtonian turbulent structures:

to orientation-dependent forcings in biological systems:

To high resolution vortices resulting from shocks:

The dataset contains a wide range of challenges from chemistry, biology, fluids, astrophysics including difficulties spanning from sharp discontinuities in the coefficient fields:

The Well was curated in collaboration with domain experts and numerical software developers specifically to provide challenging dynamics reflective of practical scientific research while staying accessible to machine learning audiences in terms of resolution and geometry.

Today along with my project co-lead @oharub and the team @PolymathicAI I'm excited to announce the release of the Well, a 15TB collection of 15+ datasets for physical simulation. Paper: Github:

RT @MilesCranmer: Some exciting @PolymathicAI news. We're expanding!!. New Research Software Engineer positions opening in Cambridge UK,….

Our team is looking for new postdocs! If you're interested in ML for physical dynamics and would like to work at the amazing @SimonsFdn in NYC, please apply.

Congrats to @liamhparker and team on pushing multi-modality forward in scientific settings in their new paper. Can't wait to see what's coming next!.

RT @cosmo_shirley: Have you all heard about ChatGPT or foundation models but want to build more than a chatbot with AI? . 🔥 We at @Polymath….

RT @cosmo_shirley: You probably all heard about #ChatGPT, but want to build more than a chatbot with AI?. 🔥 We at @PolymathicAI are buildin….

Honored to receive best paper for MPP at the @AI_for_Science @NeurIPSConf workshop with my teammates @PolymathicAI! Thanks to everyone who stopped by our poster for the great discussions and to the organizers for running such an interesting workshop! #AI4Science #NeurIPS2023.

11/ I'm at #NeurIPS2023 this week and would love to chat. I will also be presenting on Multiple Physics Pretraining with @PolymathicAI at the @ai_4_science workshop on Saturday. Hope to see you all there! . @PolymathicAI is also hiring a research engineer:

10/ This was work done with fantastic collaborators Peter Harrington and Shashank Subramanian, and my advisor Jed Brown (none of whom have Twitter accounts).

9/ As a spoiler for future work, we observed that while these sources of error also exist in attention-based models, interestingly, the models themselves are remarkably resilient to divergence.

8/ Applying our modifications directly to FourCastNet(v1), we find enormous (~8x) gains in stable rollout length for .25 degree resolution global weather forecasts without loss in accuracy, though we did not obtain the near-infinite stability we observed on simpler systems.