Autonomous robotics requires multiple concurrent functionnalities in vision, image processing, AI, path planning, mission control, data analytics from sensors (LIDAR, etc...) under real time and energy constraints. Indeed most robots use electric power with batteries and HPC consumes important energy affecting mission duration of autonomous robots. This project will extend an existing framework EFFIAROB with data analytics and performance prediction models based on an actual robot in the laboratory.
Field experiments will be conducted in the Campus of Ecole Polytechnique.
Omar Hammami omar.hammami@polytechnique.edu
data analytics and AI require high performance computing (HPC) to process large data sets in reasonnable times. Unfortunalty, these HPC requirements translate into increasing energy consumption raising questions on the use of widespread large scale data analytics and AI. In this project we propose to explore hardware accelerators dedicated to data analytics and AI using FPGA technology. Those hardware accelerators and designed from C/C++ and translated into circuits using High Level Synthesis (HLS) tools which have been proven highly energy efficient while HPC.
Intel have joined with Altera et AMD with Xilinx to offer these solutions.
The project will propose multiobjective optimization for those accelerators and actual implementation on Xilinx Board for data centers.
Omar Hammami omar.hammami@polytechnique.edu
https://calcul.math.cnrs.fr/2022-07-atelier-fpga.html
In this project we wish to implement low power HPC algorithms for social networks graphs. Typically, facebook, twitter, google among others represent data exchanges through graphs and their processing require careful load balancing in order to optimize their processing. Starting from the Stanford SNAP library the project will parallelize main functions in order to spped up processing.
Omar Hammami omar.hammami@polytechnique.edu