Generation of inference tools based on MLIR techniques for gradient management and mixed precision on diverse hardware targets

Status

Open

Scientific disciplines

Computer and Information Science

Research direction

Digital Science and Technology

Affiliate site

Rueil-Malmaison

IFP Energies Nouvelles offers an innovative PhD opportunity at the intersection of Artificial Intelligence (AI) and High-Performance Computing (HPC). This project aims to develop inference engines based on the MLIR (Multi-Level Intermediate Representation) infrastructure to optimize the performance of AI algorithms requiring precise gradient calculations, multi-precision management, and efficient execution on heterogeneous architectures (CPU, GPU, FPGA, ARM, RISC-V, SiPearl).
The PhD candidate will address key challenges such as optimizing non-linear computations with reduced latency, increased precision, and enhanced performance through multi-precision techniques. They will propose mechanisms to manage interoperability between AI models and numerical solvers in demanding industrial environments, such as subsoil modeling.
The candidate will join a multidisciplinary team and contribute to significant technological advancements in the energy industry. This project will fully leverage modern hardware architectures and optimize the compilation pipeline using MLIR.

Keywords: Generative programming, Compilation, Artificial Intelligence, Computer Science, Deep Learning

  • Academic supervisor    Dr Fabrice RASTELLO, CORSE Team, INRIA Gronoble, ORCID : 0000-0002-6589-9956
  • Doctoral School    217 - Ecole Doctorale Mathématiques, Sciences et Technologies de l'Information, Informatique, Université Grenoble Joseph Fourier 
  • IFPEN supervisor    Dr Jean-Marc GRATIEN, ORCID : 0000-0003-4721-0366
  • PhD location    IFPEN, Rueil-Malmaison, FRANCE  
  • Duration and start date    3 years, starting in the fourth quarter 2026 (Novembre 2)
  • Employer    IFPEN
  • Academic requirements    University Master degree in  Computer Science and Information System, Data sciences, Applied Mathematics   
  • Language requirements    English level B2 (CEFR)    

To apply, please send your cover letter and CV to the IFPEN supervisor indicated here below.

Contact
Encadrant IFPEN :
Dr Jean-Marc GRATIEN