Mickael SEZNEC

PhD Student

L2S, CentraleSupélec
Bât. Breguet C4.21
3 rue Joliot Curie
91190 Gif-sur-Yvette, France


Hello! I’m Mickaël Seznec, a 3rd year (CIFRE) PhD student, with Thales Research & Technology

PhD Summary

From the algorithm to the targets, optimisation flow for the real-time high performance computing on heterogeneous parallel embedded system

Thales TRT and the L2S laboratory are both tackling, in different contexts, the subject of finding optimal performance in an environment where algorithms and computing architectures tend to become more complex.
The path towards better performance for embedded applications (spatial, defence, security, public transportation…) requires the conjunction of different skillsets, often split across different teams if not different business entities. Indeed, on the one hand, applications are becoming more and more demanding in terms of computing power, and on the other hand, embedded systems require an increasingly sophisticated use of their computing resources due in particular to a heterogeneity of targets (CPU, GPU, DSP, FPGA…) and complex data communication links.
In this search for the best combination of architectures and algorithms to meet the particularly strong constraints in the field of embedded systems (real time, low energy consumption, controlled cost and of course the accuracy of the results), the frontier between expertise in the field of application and expertise in embedded systems is becoming increasingly marked.However, it reduces the space for exploring solutions likely to best optimize the application on a hardware target and can potentially prevent the achievement of the expected performances.
In this context, more in-depth optimization work at the interface between the application and embedded high-performance computing is proving to be more than necessary. 

Conference papers

An efficiency-driven approach for real-time optical flow processing on parallel hardware

Mickael Seznec, Nicolas Gac, F. Orieux, A. Sashala Naik
IEEE International Conference on Image Processing

A Study on Convolution Operator Using Half Precision Floating Point Numbers on GPU for Radioastronomy Deconvolution

Mickael Seznec, Nicolas Gac, André Ferrari, François Orieux
IEEE International Workshop on Signal Processing Systems (SiPS 2018)

Conference Posters

A new convolutions algorithm to leverage tensor cores

Mickael Seznec, Nicolas Gac, F. Orieux, A. Sashala Naik

An OpenCL design for tomographic reconstruction on FPGA. Innovate Europe Design Contest Winners

Maxime Martelli, Mickael Seznec, Nicolas Heemeryck, Nicolas Gac

Patents and software

flowpy 💾 – A python package for working with optical flows


CV – Mickaël SeznecTélécharger