PDEbench and PDEarena even use same solver to generate data (same Re, domain size, boundary for NS). Yet they arrive at contradicting conclusions. Discrepancies are likely coming from training protocol/ model tuning. Not good for #AI #PDE community to have this level of mismatch
@animaanandkumar
-
PDEbench versus PDEarena: Initial Conditions and Reproducibility
By
–
In PDEbench initial conditions are explicitly provided which is great for reproducibility @Mniepert but PDEarena has a scalar for forcing and doesn’t specify how initial conditions are generated. Opaque. @jo_brandstetter @rejuvyesh Good to fix this @eigensteve @ShuiwangJi
-
Fourier Neural Operators Achieve 10^4-10^5 Speedups Weather Forecasting
By
–
Thank you @HPCsharp @PASC_Conference https://
arxiv.org/abs/2208.05419 Our work was first to do weather forecasts at 25km spatial resolution with 10^4-10^5 speedups using Fourier neural operators. Recently we added spherical harmonics for stable long rollouts https://
arxiv.org/abs/2306.03838 -
Fourier Neural Operators Enable 25km Weather Forecasting
By
–
Thank you @PASC_Conference https://
arxiv.org/abs/2208.05419 Our work was first to do medium-range weather forecasts at 25km spatial resolution with 10^4-10^5 speedups using Fourier neural operators. Recently we added spherical harmonics for stable long rollouts https://
arxiv.org/abs/2306.03838 -
LeanDojo Enables AI Co-pilot for Theorem Proving Tasks
By
–
Thanks @AlphaSignalAI LeanDojo allows AI to be co-pilot for theorem proving. ChatGPT is not yet good at this and a small LLM with retriever works much better. We have added fully AI-written Lean proofs to ProofNet https://t.co/DYiPlvxo4N @KaiyuYang4 https://t.co/bYJbfLwwl9
— Prof. Anima Anandkumar (@AnimaAnandkumar) 28 juin 2023Thanks @AlphaSignalAI LeanDojo allows AI to be co-pilot for theorem proving. ChatGPT is not yet good at this and a small LLM with retriever works much better. We have added fully AI-written Lean proofs to ProofNet https://
github.com/zhangir-azerba
yev/ProofNet/pull/14
… @KaiyuYang4 -
AI Proves 39 New Mathematical Theorems in Lean Proof System
By
–
AI is already a co-pilot for math and theorem proving. We have proven 39 proofs in ProofNet that did not have human-written proofs in Lean. We continue to add AI-written proofs to ProofNet https://
github.com/zhangir-azerba
yev/ProofNet/pull/14
… @KaiyuYang4 -
AI Model Discovers New Mathematical Proofs in Formal Systems
By
–
Our AI model correctly discovered 33 proofs in miniF2F and 39 proofs in ProofNet of theorems that did not have existing proofs in Lean. Our proofs have helped ProofNet uncover multiple bugs in the formalization of theorem statements
-
Open Source LLM Matches OpenAI for Theorem Proving
By
–
The first open data, model and benchmark for theorem proving under MIT license. Our small LLM with retriever is as good as the closed @OpenAI model. Democratization of #AI See more details in @KaiyuYang4 thread: https://
x.com/KaiyuYang4/sta
tus/1673882824158613504?s=20
… -
Open Source LLM with Retriever Matches OpenAI in Theorem Proving
By
–
Hats off to @KaiyuYang4 for all the hard work in putting together the first open data, model and benchmark for theorem proving under MIT license. Our small LLM with retriever is as good as the closed @OpenAI model. Democratization of #AI
-
Neural Operators and AI Plus Science Conference Talk
By
–
Looking forward to my talk tomorrow on Neural Operators and AI +Science at @PASC_Conference