Dr. Chris Rackauckas 
Join us for a Dyad Modeling Livestream today - this time at 1pm ET / 10 am PT! Michael Tiller will joining us today to model a hybrid-EV powertrain!
Tune in on YouTube and send us your thoughts in the chat!
YouTube
oroikonSurrogate accuracy isn’t the same as verification. When you need credibility, analytical/manufactured solutions act like unit tests.
We’ve released an updated preprint (SIGS v3): grammar-valid symbolic candidates → latent-manifold exploration → residual-validated refinement (incl. coupled PDE systems).
For project page click here: https://oroikono.github.io/sigs-paper-site/#benchmarks
#SciML #PDE #NeuroSymbolicAI #eth-ai-center #eth #ai #ml
We’ve released an updated preprint (SIGS v3): grammar-valid symbolic candidates → latent-manifold exploration → residual-validated refinement (incl. coupled PDE systems).
For project page click here: https://oroikono.github.io/sigs-paper-site/#benchmarks
#SciML #PDE #NeuroSymbolicAI #eth-ai-center #eth #ai #ml
New fastest explicit non-stiff ODE solver? That's right, we now have something beating the pants off of the high order explicit RK methods! Check out the new symbolic-numeric optimized Taylor methods available in DifferentialEquations.jl! It uses a mix of Taylor-Mode AD, a symbolic post-processing trick, and a new order adaptivity algorithm to give a new level of performance.
See the paper: https://arxiv.org/abs/2602.04086
Your college professor teaches you "A-stable methods are required for stiff ODEs". But PSA, the most commonly used stiff ODE solvers (adaptive order BDF methods) are not A-stable. #sciml #numericalanalysis #diffeq
The most commonly used stiff ODE solver isn't A-stable?!?!?
Physics-Informed Neural Surrogates for Mesh-Invariant Modeling of High-Speed Flows at #AIAA #SciTech!
High-speed flight simulation is computationally brutal. A single CFD run can take hours on a cluster. That's fine for final validation, but not for early design exploration or real-time decision-making.
Neural surrogate that predicts aerodynamic behavior 595x faster than CFD while maintaining ~1% relative error.
Paper: https://lnkd.in/efe2Q_T9
Framework Grokkit đề xuất phương pháp "tính toán thay vì dự đoán" cho AI và KH học. Nó mã hóa toán tử liên tục, cho phép tăng độ phân giải mà không cần huấn luyện lại, giữ sai số cực thấp. Ưu điểm là chạy được trên phần cứng phổ thông vì sự phức tạp nằm ở toán học.
#AI #SciML #MachineLearning #Research #Technology #KHọc #HọcMáy #NghiênCứu #CôngNghệ
https://www.reddit.com/r/LocalLLaMA/comments/1q01el9/do_you_think_this_compute_instead_of_predict/
F. Javier RubioNew paper with J.A. Christen, just accepted in Statistical Methods in Medical Research
"Hazard-based distributional regression via ordinary differential equations"
preprint: http://arxiv.org/abs/2512.16336
R and Julia code + data: https://github.com/FJRubio67/SurvMODE
Scientific machine learning (#SciML) is not just about adding scientific information to machine learning, but it's also about making machine learning accessible to heterogeneous data.
Scientific machine learning and heterogeneous data
New livestream, #Dyad Modeling Live! In this stream we built up a thermal model of a room using #AgenticAI and added a heat pump with different control strategies and analyzed the power efficiency. Join the fun live next week! #julialang #sciml
Dyad Modeling Live: Creating a Thermal Model of a Room using Agentic AI
ANSYS /Synopsys, one of the largest simulation companies in the world, is partnering with JuliaHub in order to bring #Dyad, #Julialang, and #SciML to next level of adoption. We have many things planned. This is how research becomes reality.
