ML Researcher @proximafusion || PhD @TU_Muenchen || UQ/ML + Physics/engineering || Alumnus @IITBHU_Varanasi @CentraleNantes
Munich, Bavaria
Joined October 2015
- Tweets29
- Following141
- Followers47
- Likes91
I'll be at @NeurIPSConf 2025 in San Diego presenting our new work from @proximafusion:
ConStellaration - a dataset of QI-like stellarator boundaries + optimization benchmarks.
Poster #2211, Exhibit Hall C/D/E Dec 4, 11am-2pm
Paper: openreview.net/forum?id=NQSb…... #fusion #Al #AI4Sci
New paper alert. "Stochastic black-box optimization using multifidelity score function estimator," published in @MLSTjournal (@IOPPublishing ) w\ @mandrakeMojito @SKoutsourelakis prof. Hans Bungartz. iopscience.iop.org/article/1…
#BlackBoxOptimization #SciML #UQ #ML
Atul Agrawal retweeted
ARTICLE: From concrete mixture to structural design—a holistic optimization procedure in the presence of uncertainties
✍️@Atul_Ag07, @SKoutsourelakis (@TU_Muenchen), Erik Tamsen and J.F. Unger (@BAMResearch)
👉bit.ly/4eIo6Jo
Have an optimization problem involving a real-world physics-based simulator? The simulator is stochastic, 💵💵, black-box, multi-modal, and high dimensional? We might have a solution. Check out: hal.science/hal-04659802
(w Kislaya Ravi, @SKoutsourelakis , J.H. Bungartz)
New paper on probabilistic data driven model for turbulence closure with @SKoutsourelakis in @ElsevierPhysics Journal of Computational Physics. Open Source link: sciencedirect.com/science/ar… @TU_Muenchen #RANS #turbulence #DataDriven #SciML
Atul Agrawal retweeted
New work with @Atul_Ag07 on a probabilistic, data-driven closure model for RANS equations sciencedirect.com/science/ar…
Our submissions in @NeurIPSConf @ML4PhyS got accepted.
(1/3) When a stochastic, high-dimensional, black-box physics-based simulator is involved in constrained optimization, there are not many viable options out there. We attempt to propose one.
(collab. with @SKoutsourelakis)
Multi-fidelity Constrained Optimization for Stochastic Black Box Simulators. (arXiv:2311.15137v1 [math.OC]) ift.tt/IPnk7yp
(3/3)
We proposed a physics-based inductive bias term in a popular ML model for the unclosed term in the Reynolds-Averaged Navier-Stokes #RANS equations, leading to improved predictions.
arxiv.org/abs/2311.14576
I will be attending #NeurIPS2023 in person. Would love to chat on probabilistic machine learning + engineering/physics.
Atul Agrawal retweeted
Model uncertainty remains one the most difficult topics in #Uncertainty_Quantification. Here is our attempt to quantify it in the context of the Reynolds-Averaged Navier-Stokes model #RANS, in a physically meaningful manner that still allows probabilistic predictive estimates.
A probabilistic, data-driven closure model for RANS simulations with aleatoric, model uncertainty. arxiv.org/abs/2307.02432
Interested in learning a probabilistic, #data-driven #turbulence closure model from sparse velocity/pressure observations, while quantifying the model form error?
(Joint work with prof. @SKoutsourelakis )
#turbulence #RANS #datadriven #probabilisticmachinelearning #UQ
A probabilistic, data-driven closure model for RANS simulations with aleatoric, model uncertainty. arxiv.org/abs/2307.02432
Check out our new #JCP paper about:
Simple computational strategies for more effective physics-informed neural networks modeling of turbulent natural convection
The paper is freely available for next 50 days, click below.
lnkd.in/eArPTE-c
#turbulence #AI #PINN #fluid
Interested about efficient surrogate for turbulent thermal convection? Or ML for fluid mechanics in general 👇
Work from my time in @LisnLab ( @UnivParisSaclay)
#ML #turbulence #PINNs
Physics-aware deep neural networks for surrogate modeling of turbulent natural convection. Didier Lucor, Atul Agrawal, and Anne Sergent arxiv.org/abs/2103.03565
Atul Agrawal retweeted
Interested in using unlabeled data (i.e. just inputs) and physical constraints in constructing efficient surrogates for PDEs with high-dimensional parametric inputs in the Small-Data regime? More information in the newly-published J. of Comp. Phys. paper authors.elsevier.com/a/1cel5…
Congratulations prof @SKoutsourelakis and Sebastian.
Our paper on "Physics-aware, probabilistic model order reduction with guaranteed stability" has been accepted to #ICLR2021. Congratulations to Sebastian Kaltenbach for putting in all the hard work!
More details: openreview.net/forum?id=vyY0…
Atul Agrawal retweeted
#RAISE2020 is all set to take place from Oct 5 to 9, 2020 in a virtual avatar. The summit will host some of the world's most renowned AI leaders and decision makers - to understand how AI can be mobilised for greater social good.
Register NOW: bit.ly/3cdUBA5
#AIforAll
Thank you @Skoltech_ru @SMILESkoltech for organizing such a nice summer school on Machine Learning and accepting me :). Very interesting lectures.
.@SMILESkoltech: 2 more days with @suchisaria, @Susan_Athey, @sineadwilliamso, @PierreAlquier, @ta_broderick, @dimitrpantazis, @kelina1124, @EmtiyazKhan, @andrewgwils, @burnaevevgeny, @Jan_R_Peters, @zacharylipton Anna Goldenberg, Ruth Urner, Lior Wolf, Kun Zhang 🤗
Big whirls have little whirls,
That feed on their velocity;
And little whirls have lesser whirls,
And so on to viscosity.
- Lewis Fry Richardson
instagram.com/p/CDpAwN0FCwy/…