PhD opportunities

Polyhedral Compilation For Deep Neural Networks

Thesis proposal

Area of expertiseReal-time computer science, robotics, systems and control - Fontainebleau
Doctoral SchoolISMME - Systems Engineering, Materials, Mechanics, Energy
SupervisorMme Corinne ANCOURT
Research unitMathematics and Systems
Starting dateSeptember 1st 2021
KeywordsCompilation, Deep Neural Networks, Code optimization, Programming languages
AbstractPolyhedral compilation is a technology for automatic loop parallelization and optimization that recently raised a lot of attention in both academia and industry because of its ability to handle important artificial intelligence and machine learning codes. Current polyhedral frameworks for artificial intelligence such as Facebook’s Tensor Comprehensions or Nvidia’s Diesel apply general-purpose loop optimization techniques designed for production compilers and provided
by of-the-shelf polyhedral libraries. This research project aims at designing new polyhedral techniques specifically designed to address modern deep neural networks to exploit their properties to improve metrics such as scalability and efficiency of the generated code on a variety of target architectures. This joint project between Huawei and Mines ParisTech will support both institutions’ ongoing fundamental and applied research in cutting edge compilation for artificial intelligence frameworks.
Profile- Master 2 in computer science
- Knowledge in compilation, AI, linear programming
FundingConvention CIFRE
PDF document