Categories
seminaire passé seminar

[Seminar] October 24 2023 – Adéquation algorithme architecture pour l’accélération de méthodes d’inversion de données en grande dimension 🧑‍🏫 Nicolas GAC

🧑‍🏫 Nicolas Gac
🌎 Oct 24 2023
☕️ 10:00 AM
🏢 Maison de la Simulation, Batiment Digiteo Saclay, Salle Mandelbrot
🔗 slides

Résumé :

L’amélioration constante de la résolution des instruments parallèlement à la complexité croissante des méthodes de reconstruction basées sur des modèles de plus en plus précis, s’accompagne d’un besoin croissant en puissance de calcul. Les cartes accélératrices composées de GPU ou de FPGA sont une opportunité pour réduire l’écart technologique entre les systèmes d’acquisition et de reconstruction. Dans le contexte particulier de la résolution de problèmes inverses mal posés, mes travaux de recherche en adéquation algorithme-architecture visent à prendre en compte, en amont de la définition des méthodes, le potentiel et les limites des architectures d’accélération. Après une présentation du parallélisme offert par les architectures GPU et FPGA, le contexte algorithmique des méthodes bayésiennes et leur application en reconstruction tomographique seront présentés avec un focus sur l’accélération des opérateurs de projection/rétrojection utilisés en tomographie. Le projet ANR Dark-era portant sur l’accélération des calculs pour la radioastronomie sera ensuite présenté ; ce travail collaboratif vise à construire un outil de prototypage rapide fournissant des simulations exascales des futurs serveurs HPC nécessaires au traitement « temps réel » du flux de données massives du radiotélescope SKA. Enfin, des perspectives sur ces travaux seront exposées.

Categories
highlight job thesis

[JOB] 👩‍💻🧑‍💻Optimization framework for large deep neural networks

[FR] Plateforme logicielle d’optimisation pour réseaux de neurones profonds de très grande taille

Doctoral Domain : Computer Science

Funding : CIFRE schema, hired in French CDD (36 months) by Huawei France

Work Places : Huawei Paris Research Center (Boulogne-Billancourt) and Maison de la Simulation (University Paris-Saclay)

Industrial Supervision : Dr. Chong Li and Prof. Denis Barthou, Huawei Paris Distributed and Parallel Technologies Lab

Academic Supervision : Prof. Nahid Emad, Université Versailles Saint-Quentin-en-Yvelines

Keywords : Massively Parallelism, Linear Algebra, Performance Optimisation, Deep Learning, Large Models

Categories
cdi

[JOB] Permanent Researcher Position

System software for in situ HPC/AI coupling

Maison de la Simulation (MdlS) recruits a permanent CEA researcher (ingénieur chercheur permanent) to reinforce its “Science of Computing” (SoC) team.

In order to apply, please send a resume, cover letter, references and support letters to info@maisondelasimulation.fr. You can use the same email address for requests of information about the position. Applications will be evaluated from May the 15th 2023 and until the position is filled.

Context

Maison de la Simulation is a joint research and engineering laboratory of CEA, CNRS, Université Paris-Saclay and Université de Versailles Saint-Quentin-en-Yvelines localized on the plateau de Saclay campus next to Paris. It specializes in high-performance computing. The “Science of Computing” team conducts research, builds expertise and engineers tools in domains underlying HPC: computer science and applied mathematics.

The team co-leads the 5-years French ExaDoST project, part of NumPEx, that will design the software stack for data handling on the upcoming French and potentially European Exascale super-computers.

Mission

During the first five years, the selected candidate will contribute to the ExaDoST project. They will conduct research and design tools for data handling at Exascale, working both on the user facing API and on system software for in situ HPC/AI coupling. They will work in close collaboration with the other members of the project to design a modular software stack that should be used on the upcoming French Exascale supercomputer as well as more globally on French, European, and worldwide supercomputers.

To achieve these goals, the candidate will take part in the recruitment and management of the group of temporary engineer and young researchers dedicated to this project. They will leverage the expertise and tools already developed in the team, including the PDI and Deisa libraries.

As time evolves, the selected candidate will develop their own research and projects in the domains of the team. At the conclusion of ExaDost, they will have a large latitude to direct their research toward the directions they feel would serve the team and laboratory best.

Main activities

The candidate will:

  • conduct their research in the domain of runtime and system software for in situ analytics and HPC/AI coupling,
  • take part in the design and implementation of libraries and tools that make the results of this research available to users of HPC in production,
  • take part in the NumPEx project, and related activities of management and reporting,
  • participate to the management of temporary engineers and young researchers.

Skills and competencies

The successful candidate will hold a PhD thesis in computer science or in another field with a strong experience in computer science. They should master the following skills and competencies:

  • good knowledge of HPC parallel architectures, operating systems, and application programming,
  • knowledge of the design of existing system tools and libraries for HPC data handling (IO, in situ processing, coupling, checkpoint writing, etc.),
  • good programming skills in C++, and associated developments tools (Cmake, git, etc.),
  • good programming skills in python, and associated ecosystem,
  • autonomy, interest for team-work in an international context, leadership.

Salary and advantages

CEA “ingénieur-chercheur” positions offer a very competitive salary in French research ecosystem, indexed on diplomas and experience. In addition, the selected candidate will benefit from many advantages:

  • possibility to leverage existing collaborations of MdlS in Europe, US and Japan as well as international conferences with support for travel,
  • up to 3 days of remote work per week,
  • reimbursement of public transport costs (75%) and subsidized catering service,
  • included pension plan and health coverage (French social security plus CEA-specific insurance),
  • 9 full weeks of holidays per year.
Categories
member

Members

Director

Pascal TREMBLIN (CEA senior research scientist)

Executive assistant

Valérie BELLE

Permanent members

Science by Computing Team
  • Team Leader:
    Karim HASNAOUI (CNRS research engineer)
  • Edouard AUDIT (CEA senior research scientist)
  • Daniel BORGIS (CNRS senior research scientist)
  • Mathieu LOBET (CEA research scientist)
  • Riccardo SPEZIA (CNRS senior research scientist)
  • Yushan WANG (CEA research scientist)
  • Charles PROUVEUR (CNRS research engineer)
Science of Computing Team
  • Team Leader:
    Julien BIGOT (CEA research scientist)
  • Victor ALESSANDRINI (CNRS research scientist emeritus)
  • Yuuichi ASAHI (CEA research scientist)
  • Benoît MARTIN (CEA research scientist)
  • Simplice DONFACK (CNRS research engineer)
  • Thomas DUFAUD (UVSQ associate professor)
  • Nahid EMAD (UVSQ full professor)
  • Martial MANCIP (CNRS research engineer)
  • Thomas PADIOLEAU (CEA research scientist)
Project Officer
  • Julien THELOT

Fixed-term members

Postdocs
  • Paul REGNAULT
  • Simon EL-OUAFA
Engineers
  • Juan Jose SILVA CUEVAS
  • Salem KHELLAL
  • Hiba TAHER
  • Paul ZEHNER
  • Thierry ANTOUN
PhD students
  • Rémi BOURGEOIS
  • Jean-Marc LUDE
  • Aymeric MILLAN
  • François-Xavier MORDANT
  • Quentin PETIT
  • Kevin PEYEN
  • Mathilde POVEDA
  • Lou ROUSSEL-HARD
  • Pascal WANG
  • Zineb ZIANI
Trainees
  • Merieme BOURENANE
  • Benjamin GERBE
  • Melissa LARBI
  • Florent NYBERG
  • Pierre-Antoine RACLIUS
  • Sarah SAADOUNI

Former members