Avatar

Mathieu Léonardon

Associate Professor in Electronics

IMT Atlantique

Biography

I conduct research at the IMT Atlantique in Brest on hardware and software implementations of signal processing and AI algorithms. I teach computer engineering and digital electronics.

My PhD thesis focused on the implementation of polar codes decoders. I proposed the fastest software implementation of the Adaptive SC List decoding algorithm to date. This implementation is integrated in the AFF3CT toolbox to which I actively contribute.

I currently focus on efficient hardware and software implementations of Neural Networks, aiming at low latency and energy efficiency, through multiple industrial collaborations, and as the coordinator of a JCJC ANR project, ProPruNN. We also recently won the AMD Open Hardware Competition with the PEFSL project, a pipeline for the training, compilation, hardware synthesis and deployment of a few-shot learning application on an FPGA SoC.

Interests

  • Neural Networks Compression
  • Embedded Electronics
  • Channel Coding
  • HPC

Education

  • PhD in Electronics, 2018

    Polytechnique Montréal

  • PhD in Electronics, 2018

    University of Bordeaux

  • MEng in Embedded Electronics, 2015

    Enseirb-Matmeca, Bordeaux INP

PEFSL

A Pipeline for Embedded Few-Shot Learning

A modular pipeline for the training, compilation, hardware synthesis and deployment of a few-shot learning application on an FPGA SoC.

Star

Source code on GitHub

AFF3CT

A Fast Forward Error Correction Toolbox

Simulate high-throughput communication chains.

Star

Source code on GitHub Website

Latest release

Recent Publications

Federated learning compression designed for lightweight communications

Federated Learning (FL) is a promising distributed method for edge-level machine learning, particularly for privacysensitive …

Pipelined Architecture for a Semantic Segmentation Neural Network on FPGA

Many machine vision tasks like urban sceneunderstanding rely on machine learning, and more specifically deep neural networks to provide …

Compression de réseaux de neurones pour l'apprentissage fédéré

Federated Learning (FL) is a promising distributed method for edge-level machine learning, particularly for privacysensitive …

Leveraging Structured Pruning of Convolutional Neural Networks

Structured pruning is a popular method to reduce the cost of convolutional neural networks, that are the state of the art in many …

Contact

  • +33 2 29 00 13 84
  • 655 Avenue du Technopôle, Plouzané, 29280, FRANCE
  • Enter Building K2 and take the stairs to Office K2 216A
  • Skype Me