Logo Sciencesconf

HACE - workshop on HPC/AI hybridization

Institut de Recherche en Informatique de Toulouse (IRIT), Toulouse, France

May 26th - 28th, 2026

Location: IRIT – 118 Route de Narbonne, Cours Rose Dieng-Kuntz, F-31062 TOULOUSE CEDEX 9

Auditorium: JACQUES HERBRAND

 

This event belongs to the thematic trimester "HPC and AI convergence at the exascale era" organized by AISSAI Center and PEPR NumPEx in collaboration with ANITI AI cluster.

 

HACE workshop in Toulouse (May 26–28 2026) is the technical follow‑up of the opening conference SCOPE on “HPC and AI convergence at the Exascale era that took place in Paris in March 2026. The workshop will gather researchers from applied mathematics, scientific computing, AI, and HPC engineering to explore how hybrid AI–HPC workflows can deliver robust, scalable solvers for complex physical and industrial applications. By articulating methodological advances (PINNs, Neural Operators, diffusion and foundation models) with large‑scale implementations and real application case‑studies, HACE aims at identifying concrete co‑design strategies between numerical methods, learning architectures, and exascale systems. The format combines keynote talks, focused application talks, and hands‑on / technical sessions to foster exchanges between communities and to help participants build reusable patterns for hybrid AI–HPC workflows in their own domains.

Registration

Participation is free of charge, but registration is mandatory.

All the talks will be at the IRIT

Please register here.

 

Keynote speakers

 

Olivier Beaumont - Inria

Quentin Bertrand - Inria

Youngsoo Choi - Lawrence Livermore National Laboratory

Alban Farchi - École des Ponts Paris Tech / CEREA

Alena Kopanikacova - Toulouse INP

Nikola Kovachki - NYU, Nvidia

Jean Kossaifi - Nvidia

Philipp Krah - CEA

Fanny Lehmann - ETH Zürich

Oriol Lehmkuhl - BSC

Andreas Lintermann - FZ Jülich

Benoît Malezieux - CNRS

Edouard Oyallon - CNRS

Bruno Raffin - Inria

Laure Reynaud - Météo-France

Nilo Schwencke - Université Paris-Saclay

Luciano Drozda - CERFACS

Alex Zinenko - AMD

Mathilde Mougeot - ENSIIE 

Program

Tuesday, May 26th, 2026 - 8:15 AM to 6:30 PM CEST (Day 1)

8.15 AM - 9.00 AM (CEST) Registration of participants and welcome coffee ☕.

9:00 AM - 9:15 AM (CEST) Welcome and opening words (Presentation of the AISSAI semester)

Speaker: Jerôme Bobin

9.15 AM - 12.40 PM (CEST) Morning session: Deep-dive into AI-based surrogates to solve PDEs

This first morning session is devoted to AI‑based surrogates for solving partial differential equations, with a focus on Physics‑Informed Neural Networks (PINNs) and Neural Operator architectures. It will cover recent theoretical and algorithmic advances, discuss training and generalization issues for complex multi‑scale systems, and explore how these models interface with traditional solvers on modern HPC platforms. Invited contributions will present PINN‑based approaches, Neural Operator methods, and selected applications ranging from plasma physics to fluid dynamics, highlighting both performance gains and current limitations in terms of accuracy, stability, and scalability.

Chair: Emmanuel Franck

  • 9:15 AM - 10:10 AM (CEST) 45 min for a general talk on PINNs + 10 min of Q&A - Alena Kopanikacova (IRIT - France) 
  • 10:10 AM - 10:50 AM (CEST) Applications talks - Nilo Schwencke (LISN - France) - (30 min of presentation + 10 min of Q&A)

10:50 AM - 11:20 AM (CEST) Coffee break ☕

  • 11:20 AM - 12:00 PM (CEST) Applications talks - Nikola Kovachki (Nvidia - US) - (30 min of presentation + 10 min of Q&A)
  • 12:00 PM - 12:40 PM (CEST) Applications talks - Jean Kossaiffi (Nvidia - US) (30 min of presentation + 10 min of Q&A)

12:40 PM - 2 PM (CEST) Lunch Break 🍴

 

2:00 PM - 5:20 PM (CEST) Afternoon session - HPC/AI hybridization         

The second afternoon focuses on the hybridization of HPC and AI, from system‑level integration to programming models and software stacks. Presentations will cover scalable training and inference on heterogeneous architectures, workflow orchestration, and co‑design of algorithms and runtime systems for coupled AI‑physics simulations. Talks on AI–HPC coupled workflows, compiler and IR technologies (e.g. MLIR‑based approaches), and large‑scale production environments will highlight concrete patterns for deploying hybrid applications at scale, including collaborations with industrial and vendor partners.

Chair: Alfredo Buttari

  • 2:00 PM - 2:40 PM (CEST) AI/HPC coupled workflows - Bruno Raffin (INRIA - France) - (30 min of presentation + 10 min of Q&A)
  • 2:40 PM - 3:20 PM (CEST) Benoît Malézieux (CNRS - France) - (30 min of presentation + 10 min of Q&A)
  • 3:20 PM - 4:00 PM (CEST) MLIR compiler infrastructure for scientific computing - Alex Zinenko (AMD - France) - (30 min of presentation + 10 min of Q&A) 

4:00 PM - 4:20 PM (CEST) Coffee break ☕

  • 4:20 PM - 5:00 PM (CEST) Training at scale - Edouard Oyallon (CNRS/Sorbonne University - France) - (30 min of presentation + 10 min of Q&A) 
  • 5:00 PM - 5:40 PM (CEST) Training at scale - Olivier Beaumont (INRIA - France) - (30 min of presentation + 10 min of Q&A) 

5.40 PM - 6.30 PM (CET) COCKTAIL & APERITIF 🍸🍹

 

Wednesday, May 27th, 2026 - 8:15 AM to 4:20 PM CEST (Day 2)

8.15 AM - 9.00 AM (CEST) Registration of participants and welcome coffee ☕.

9.00 AM - 12.40 PM (CEST) Morning session - From Generative models to Foundation models

This morning session will explore generative models and foundation models for scientific computing, with an emphasis on diffusion models and foundation models for scientific computing. A focus will be given on the mathematical foundations as well as their applications to real physical cases.

Chair: Benoît Malézieux

  • 9:00 AM - 9:45 AM (CEST) 45 min for a general talk on Diffusion models + 10 min of Q&A - Quentin Bertrand (INRIA - France)
  • 9:45 AM - 10:25 AM (CEST) Fanny Lehmann (ETH AI Center - Switzerland)(30 min of presentation + 10 min of Q&A)
  • 10:25 AM - 11:05 AM (CEST) Youngsoo Choi (LLNL’s CASC group - US)(30 min of presentation + 10 min of Q&A)

11:05 AM - 11:20 AM (CEST) Coffee break ☕

 

11:20 AM - 12:40 PM (CEST) -  PANEL DISCUSSION - Benchmarks AI-based methods, from PINNs to FMs.  

The panel will address benchmarking of AI‑based methods—from PINNs to foundation models—covering metrics, datasets, and best practices to compare learned surrogates with traditional HPC solvers in a scientifically meaningful way.

Moderator: Pascal Tremblin

Speakers: Fanny Lehmann, Youngsoo Choi, Bruno Raffin, Nikola Kovachki and Edouard Oyallon. 

  • 11.20 PM - 12.20 PM Round table discussion - Theme: Benchmarks AI-based methods, from PINNs to FMs. 
  • 12.20 PM - 12.40 PM Q&A session between the panellists and the audience following the roundtable discussion

12:40 PM - 2:00 PM - Lunch Break 🍴

 

2.00 PM - 4.20 PM (CEST) Afternoon session - Applications

This afternoon is dedicated to application‑driven talks showcasing hybrid AI–HPC workflows in climate and weather, computational fluid dynamics, and industrial use‑cases. Speakers will present how AI surrogates and learned components are embedded within operational or near‑operational pipelines.

Chair: Jerôme Bobin

  • 2:00 PM - 2:40 PM (CEST) Laure Raynaud (Meteo-France - France) (30 min of presentation + 10 min of Q&A)
  • 2:40 PM - 3:20 PM (CEST) Alban Farchi (ECMWF - UK) (30 min of presentation + 10 min of Q&A)
  • 3:20 AM - 4:00 PM (CEST) Andreas Lintermann (Forschungszentrum Jülich - Germany) (30 min of presentation + 10 min of Q&A)

4:00 PM - 4:30 PM - Coffee break ☕

  • 4:30 PM - 5:10 PM (CEST) Luciano Drozda (CERFACS - France) (30 min of presentation + 10 min of Q&A)
  • 5:10 PM - 5:50 PM (CEST) Mathilde Mougeot (ENSIIE - France) (30 min of presentation + 10 min of Q&A)            

 

Thursday, May 28th, 2026 - 8:15 AM to 1:00 PM CEST (Day 3)

8.15 AM - 9.00 AM (CEST) Registration of participants and welcome coffee ☕.

9.00 AM - 12.10 PM (CEST) Morning session - Applications

This morning is a follow-up to the first session dedicated to application‑driven talks showcasing hybrid AI–HPC workflows. It will end with a panel session that will discuss the challenges of AI for scientific computing

Chair: Luciano Drozda

  • 9:00 AM - 9:40 AM (CEST) Philipp Krah (IRFM-CEA - France) (30 min of presentation + 10 min of Q&A)
  • 9:40 AM - 10:20 AM (CEST) Oriol Lehmkuhl (BSC - Spain) (30 min of presentation + 10 min of Q&A)

10:20 AM - 11:00 AM Coffee break ☕       

11:00 AM - 1:00 PM (CEST) -  PANEL DISCUSSION - Application/Technical talks

This panel discussion aims at discussing how AI is employed in scientific computing in application domains, its promises and challenges.

Moderator: Luciano Drozda

Speakers: Oriol Lehmkuhl, Laure Raynaud, Alban Farchi, and Andreas Lintermann.

  • 11:00 AM - 12:00 AM Round table discussion - Theme: Application/Technical talks
  • 12:00 AM - 1:00 PM Q&A session between the panellists and the audience following the roundtable discussion

1:00 PM - 1:10 Closing words

 

This event is orgnized by PEPR NumPEx and AISSAI Center

 

 

Practical informations 

- 

Wifi access is available through Eduroam

  

Organized by

FR2030_Nume_rique_pour_l_exascale_3.jpg

logo_AISSAI_250px_1.png

 

 

In collaboration with

Loading... Loading...