Machine Learning R&D Engineer

Kinetix

Biography

I’m a software engineer that acts as a machine learning researcher. Or maybe the other way round. I’m interested in deep learning and its applications, from research to production.

Interests
  • Deep learning
  • Machine learning software development
  • Cloud computing
Education

  • PhD in Artificial Intelligence, 2021

    INSA Rouen, Normandy University, France

  • MSc in Software Engineering (1st/43), 2017

    University of Orléans, France

  • BSc in Computer Science (2nd/64), 2015

    University of Orléans, France

Experience

 
 
 
 
 
R&D Engineer
Kinetix
May 2022 – Present Paris, France
R&D engineer in the Real-Time squad. Improving our systems’ wall-clock performance and developing a real-time version of the motion capture pipeline. Faster algorithms, model pruning, knowledge distillation, cloud architecture, software optimization.
 
 
 
 
 
R&D Engineer
LITIS, Normandy University
Jun 2021 – Apr 2022 Rouen, France
R&D engineer, Research chairs RAIMo and ICUB for safety and security of autonomous driving systems. Industrial partnerships with Peugeot S.A. (Stellantis) and IRT SystemX.
 
 
 
 
 
Ph.D Student
INSA Rouen, LITIS, Normandy University
Sep 2017 – Apr 2021 Rouen, France
Ph.D student on deep generative models. Multi-obective training of Generative Adversarial Nets. Applications to scientific data, most notably geostatistical data.
 
 
 
 
 
Research Intern
LITIS, Normandy University
Apr 2017 – Sep 2017 Rouen, France
Deep Learning for offline handwritten text recognition on mobile devices with Convolutional LSTMs. Industrial partnership with Hamelin SAS for the Oxford brand of office supplies. Development of a prototype recognition system with TensorFlow
 
 
 
 
 
Research Assistant
LIFO, University of Orléans
Apr 2015 – Jul 2015 Orléans, France
Theoritical works on the SMART (Small Minimal Aperiodic Reversible Turing machine) computation model. Development of a cross-platform application for researchers for the visualisation of the SMART machine, with OpenFL.

Recent Publications

Physically-admissible polarimetric data augmentation for road-scene analysis
Polarimetric imaging, along with deep learning, has shown improved performances on different tasks including scene analysis. However, …
PDF Cite Code DOI
PhD Thesis: Auxiliary Tasks for the Conditioning of Generative Adversarial Networks
In this thesis, we explore the conditioning of Generative Adversarial Networks, that is influencing the generation process in order to control the content of a generated image. We focus on conditioning through auxiliary tasks, that is we explicitly implement additional objective to the generative model to complement the initial goal of learning the data distribution.First, we introduce generative modeling through several examples, and present the Generative Adversarial Networks framework.
PDF Cite Slides
Pixel-wise Conditioned Generative Adversarial Networks for Image Synthesis and Completion
We address the task of learning effective generative adversarial networks when only very few pixel values are known beforehand.
PDF Cite Code
Gradient-based deterministic inversion of geophysical data with generative adversarial networks: Is it feasible?
We study the use of GANs for gradient-based deterministic inversion in Geophysics and highlight the associated difficulties caused by the nonlinear GAN transform.
PDF Cite DOI
Pixel-wise Conditioning of Generative Adversarial Networks
We investigate adding a regularization term to the conditioning of GANs to deal with cases where a small subset of the image is known beforehand.
PDF Cite Code Slides
See all publications

Teachings

INSA Rouen

  • Computer science 1:
    • Introduction to algorithmics
    • Introduction to programming in Pascal
  • Applied statistics for data science:
    • Statistics, data analysis and visualisation, regressions, dimensionality reduction, testing
    • Practical lessons and projects in numpy, matplotlib, pandas

Interns

Generative models for image generation under physical constraints

El Hadji Brane Seck

June to September 2019

INSA Rouen, LITIS lab, France

Developped a CycleGAN-like model for generating physically-realistic polarimetric images under optical constraints. Work eventually published in CVIU (2022)

Conditional generative models for image generation under pixel-wise constraints

Lucas Anquetil, Pierre Lopez

June to August 2019

INSA Rouen, LITIS lab, France

Model selection and hyperparameter tuning for pixel-wise conditionned GANs. Work published in Neurocomputing (2020).

Hobbies

Music

Learning and playing bass guitar

Lockpicking

Amateur lockpicker and locksmith

Homebrewing

Hobbyist PlayStation Vita homebrew developer