Chris Rackauckas
@chrisrackauckas.bsky.social
Lead dev of SciML org, VP of Modeling and Simulation @JuliaHub, Director of Scientific Research @PumasAI, and Research Staff
@mit_csail. #julialang #sciml
@mit_csail. #julialang #sciml
BLIS is a really slow BLAS lol
August 4, 2025 at 6:23 PM
BLIS is a really slow BLAS lol
New paper on fitting #neuroscience models to neuroimaging and electrophysiological data! We show that the previous techniques within the field (Spectral DCM) can be greatly improved using automatic differentiation, #julialang #sciml, ModelingToolkit, and more.
July 25, 2025 at 11:33 AM
New paper on fitting #neuroscience models to neuroimaging and electrophysiological data! We show that the previous techniques within the field (Spectral DCM) can be greatly improved using automatic differentiation, #julialang #sciml, ModelingToolkit, and more.
We can learn dynamical systems that capture the correct statistical properties of the original system, and use symbolic regression to sparsify the results to predicted governing equations. Use case: learning neural mass models, i.e. models of average behaviors of large brain regions
July 9, 2025 at 10:29 AM
We can learn dynamical systems that capture the correct statistical properties of the original system, and use symbolic regression to sparsify the results to predicted governing equations. Use case: learning neural mass models, i.e. models of average behaviors of large brain regions
However, we can force the perturbed model to be non-chaotic, making the fitting process orders of magnitude easier. It also gives a smooth loss landscape with less or no bad local minima.
July 9, 2025 at 10:29 AM
However, we can force the perturbed model to be non-chaotic, making the fitting process orders of magnitude easier. It also gives a smooth loss landscape with less or no bad local minima.
So how did we overcome this? By making a non-chaotic training of course! We instead fit a perturbed model with the curious behavior that it retains the same global optima, i.e. the correct fitting parameters of the perturbed model is also a correct fitting model of the original model.
July 9, 2025 at 10:29 AM
So how did we overcome this? By making a non-chaotic training of course! We instead fit a perturbed model with the curious behavior that it retains the same global optima, i.e. the correct fitting parameters of the perturbed model is also a correct fitting model of the original model.
Our new manuscript shows how to extend automated model discovery and universal differential equations to chaotic systems in #neuroscience using a trick from control literature known as the Prediction Error Method (PEM)!
Check out the paper at: arxiv.org/abs/2507.03631
#sciml #ai4science
Check out the paper at: arxiv.org/abs/2507.03631
#sciml #ai4science
July 9, 2025 at 10:29 AM
Our new manuscript shows how to extend automated model discovery and universal differential equations to chaotic systems in #neuroscience using a trick from control literature known as the Prediction Error Method (PEM)!
Check out the paper at: arxiv.org/abs/2507.03631
#sciml #ai4science
Check out the paper at: arxiv.org/abs/2507.03631
#sciml #ai4science
It is built upon the #julialang #sciml ecosystem, meaning the generated code is accessible and easily re-targetable. You can write your own backends, force the usage of #gpu acceleration, do differentiable programming, etc.
June 18, 2025 at 10:30 AM
It is built upon the #julialang #sciml ecosystem, meaning the generated code is accessible and easily re-targetable. You can write your own backends, force the usage of #gpu acceleration, do differentiable programming, etc.
Dyad comes with #VSCode extensions to seamlessly perform static analyses for furthering model correctness by ensuring the physics is correct before code generation.
June 18, 2025 at 10:30 AM
Dyad comes with #VSCode extensions to seamlessly perform static analyses for furthering model correctness by ensuring the physics is correct before code generation.
With modeling heat pumps and dehumidifiers, we were able to show that the latest boundary value problem (#BVP) solvers in #julialang #sciml outperform the Fortran wrapped bvp_solver of SciPy and the native bvp4c/5c solvers of MATLAB.
Published article here: authors.elsevier.com/c/1lHcein8Vr...
Published article here: authors.elsevier.com/c/1lHcein8Vr...
June 18, 2025 at 10:20 AM
With modeling heat pumps and dehumidifiers, we were able to show that the latest boundary value problem (#BVP) solvers in #julialang #sciml outperform the Fortran wrapped bvp_solver of SciPy and the native bvp4c/5c solvers of MATLAB.
Published article here: authors.elsevier.com/c/1lHcein8Vr...
Published article here: authors.elsevier.com/c/1lHcein8Vr...
New blackbox global optimization benchmarks shows that the #julialang BlackBoxOptim.jl (BBO) outperforms the classic C-based NLopt on a wide array of difficult optimization problems.
See the benchmarks for more details: docs.sciml.ai/SciMLBenchma...
#optimization #machinelearning #C #python
See the benchmarks for more details: docs.sciml.ai/SciMLBenchma...
#optimization #machinelearning #C #python
May 31, 2025 at 10:54 AM
New blackbox global optimization benchmarks shows that the #julialang BlackBoxOptim.jl (BBO) outperforms the classic C-based NLopt on a wide array of difficult optimization problems.
See the benchmarks for more details: docs.sciml.ai/SciMLBenchma...
#optimization #machinelearning #C #python
See the benchmarks for more details: docs.sciml.ai/SciMLBenchma...
#optimization #machinelearning #C #python
Earn money working on open source software #oss! New project just posted: help make wrappers to connect Symbolics.jl to SymPy. $300 bounty. Information for signing up for the #SciML small grants program are contained in the link:
sciml.ai/small_grants...
#julialang #python #symbolics #sympy #ode
sciml.ai/small_grants...
#julialang #python #symbolics #sympy #ode
May 29, 2025 at 9:20 AM
Earn money working on open source software #oss! New project just posted: help make wrappers to connect Symbolics.jl to SymPy. $300 bounty. Information for signing up for the #SciML small grants program are contained in the link:
sciml.ai/small_grants...
#julialang #python #symbolics #sympy #ode
sciml.ai/small_grants...
#julialang #python #symbolics #sympy #ode
Using higher order automatic differentiation to improve stiff ODE solvers? Using a third order Newton-like method (Halley's) inside the #sciml #julialang ODE solvers with Taylor-mode AD, ~25% faster.
arxiv.org/abs/2501.16895
arxiv.org/abs/2501.16895
January 29, 2025 at 10:15 AM
Using higher order automatic differentiation to improve stiff ODE solvers? Using a third order Newton-like method (Halley's) inside the #sciml #julialang ODE solvers with Taylor-mode AD, ~25% faster.
arxiv.org/abs/2501.16895
arxiv.org/abs/2501.16895
We demonstrate this on a turbofan jet engine, achieving 0.1% relative error through an active learning process. This is one of the JuliaHub #julialang demonstrations from #scitech showcasing the advancements of industrialization of #SciML in #JuliaSim
January 16, 2025 at 8:13 AM
We demonstrate this on a turbofan jet engine, achieving 0.1% relative error through an active learning process. This is one of the JuliaHub #julialang demonstrations from #scitech showcasing the advancements of industrialization of #SciML in #JuliaSim
This new Radau method achieves state of the art performance on many highly stiff ODEs when high accuracy is needed, outperforming the classic #fortran and #c++ libraries as well as a lot of recent #julialang #sciml improvements!
December 20, 2024 at 2:38 AM
This new Radau method achieves state of the art performance on many highly stiff ODEs when high accuracy is needed, outperforming the classic #fortran and #c++ libraries as well as a lot of recent #julialang #sciml improvements!
Outperforms the classic Hairer Fortran implementation of radau by about 2x across the board!
December 20, 2024 at 2:38 AM
Outperforms the classic Hairer Fortran implementation of radau by about 2x across the board!
New version of a very good ODE solver today! IRKGaussLegendre released a SIMD and multithreaded mode. 16th order Implicit Runge-Kutta integrator IRKGL16 for non-stiff symplectic equations which require high accuracy.
For more benchmarks, see the github.com/SciML/IRKGau...
#julialang #sciml #ode
For more benchmarks, see the github.com/SciML/IRKGau...
#julialang #sciml #ode
November 9, 2024 at 3:20 PM
New version of a very good ODE solver today! IRKGaussLegendre released a SIMD and multithreaded mode. 16th order Implicit Runge-Kutta integrator IRKGL16 for non-stiff symplectic equations which require high accuracy.
For more benchmarks, see the github.com/SciML/IRKGau...
#julialang #sciml #ode
For more benchmarks, see the github.com/SciML/IRKGau...
#julialang #sciml #ode
Wanted to get paid to contribute to open source? Help update scientific models in benchmarks of equation solvers (differential equations, nonlinear optimization, physics-informed neural networks) and collect some dough while doing so! See the link below for details.
#julialang #pinn #sciml
#julialang #pinn #sciml
November 4, 2024 at 11:01 AM
Wanted to get paid to contribute to open source? Help update scientific models in benchmarks of equation solvers (differential equations, nonlinear optimization, physics-informed neural networks) and collect some dough while doing so! See the link below for details.
#julialang #pinn #sciml
#julialang #pinn #sciml
PumasAI named Best Clinical Pharmacology Technology Firm by the 9th Annual Biotechnology Awards! This demonstrates the power of translating #julialang #sciml to industrial practice, building a new foundation of clinical pharmacology.
May 10, 2024 at 11:40 PM
PumasAI named Best Clinical Pharmacology Technology Firm by the 9th Annual Biotechnology Awards! This demonstrates the power of translating #julialang #sciml to industrial practice, building a new foundation of clinical pharmacology.
Differentiable Metropolis-Hastings: differentiate through Bayesian estimation to optimize models towards achieving desired probabilistic outcomes, with implementation in #julialang (#sciml)
For more information, see arxiv.org/abs/2306.07961
For more information, see arxiv.org/abs/2306.07961
April 14, 2024 at 4:07 PM
Differentiable Metropolis-Hastings: differentiate through Bayesian estimation to optimize models towards achieving desired probabilistic outcomes, with implementation in #julialang (#sciml)
For more information, see arxiv.org/abs/2306.07961
For more information, see arxiv.org/abs/2306.07961
New structural identifiability analysis features: automatically reparameterize an ODE system to find the best way to make a system easier to learn with #julialang #SciML differentiable programming! For more, leave a star at github.com/SciML/Struct... and check out the tutorial
April 13, 2024 at 11:42 AM
New structural identifiability analysis features: automatically reparameterize an ODE system to find the best way to make a system easier to learn with #julialang #SciML differentiable programming! For more, leave a star at github.com/SciML/Struct... and check out the tutorial
#julialang GPU-based ODE solvers which are 20x-100x faster than those in #jax and #pytorch? Check out the paper on #sciml DiffEqGPU.jl. Instead of relying on high level array intrinsics that #machinelearning libraries use, it use a direct kernel generation!.
www.sciencedirect.com/science/arti...
www.sciencedirect.com/science/arti...
December 25, 2023 at 1:58 PM
#julialang GPU-based ODE solvers which are 20x-100x faster than those in #jax and #pytorch? Check out the paper on #sciml DiffEqGPU.jl. Instead of relying on high level array intrinsics that #machinelearning libraries use, it use a direct kernel generation!.
www.sciencedirect.com/science/arti...
www.sciencedirect.com/science/arti...
#julialang #sciml for perfume engineering. Mixing physical models with machine learned quantities in order to predict and classify odors.
chemrxiv.org/engage/chemr...
chemrxiv.org/engage/chemr...
December 23, 2023 at 7:21 PM
#julialang #sciml for perfume engineering. Mixing physical models with machine learned quantities in order to predict and classify odors.
chemrxiv.org/engage/chemr...
chemrxiv.org/engage/chemr...
New open source tool from JuliaHub: static code analysis to prove that a #julialang code is allocation-free. Use this to ensure that codes are safe for real-time applications, such as how we use it for JuliaSim to analyze #SciML control codes!
github.com/JuliaLang/Al...
github.com/JuliaLang/Al...
November 18, 2023 at 6:02 PM
New open source tool from JuliaHub: static code analysis to prove that a #julialang code is allocation-free. Use this to ensure that codes are safe for real-time applications, such as how we use it for JuliaSim to analyze #SciML control codes!
github.com/JuliaLang/Al...
github.com/JuliaLang/Al...
Direct results for the GPU performance of #julialang #sciml vs #jax vs #pytorch.
To use the code, check out the documentation at docs.sciml.ai/DiffEqGPU/st...
To use the code, check out the documentation at docs.sciml.ai/DiffEqGPU/st...
November 18, 2023 at 5:02 PM
Direct results for the GPU performance of #julialang #sciml vs #jax vs #pytorch.
To use the code, check out the documentation at docs.sciml.ai/DiffEqGPU/st...
To use the code, check out the documentation at docs.sciml.ai/DiffEqGPU/st...
Hey #julialang, trying the new app! First thing to share is I did a #sciml #systembiology #computationalbiology podcast where I talked about Julia, SciML, and Catalyst.jl.
See the episode: lnkd.in/eGTWnBit
See the episode: lnkd.in/eGTWnBit
November 11, 2023 at 10:17 PM
Hey #julialang, trying the new app! First thing to share is I did a #sciml #systembiology #computationalbiology podcast where I talked about Julia, SciML, and Catalyst.jl.
See the episode: lnkd.in/eGTWnBit
See the episode: lnkd.in/eGTWnBit