Julien Girard-Satabin

PhD student on software safety and deep learning




I am a PhD student working on software safety applied to artificial intelligence. My research interests include abstract interpretation and symbolic propagation, SAT/SMT solving applied to deep neural networks. I am also interested in the formulation and formalization of specifications for programs using audio and visual inputs.


  • Software Safety
  • Formal Verification Methods
  • Artificial Intelligence
  • Deep learning


  • PhD in Artificial Intelligence, 2018-2021

    Université Paris-Sud

  • « Diplôme d'ingénieur » (MEng french equivalent), 2014-2018

    ENSTA Paris

  • « Classes préparatoires », 2012-2014

    Lycée Massena, Nice



I am proficient in Python and usual scientific libraries (numpy, scikit, pytorch)

Functional programming

I am writing a medium-sized project in OCaml using the dune build manager. I also have a basic knowledge of Haskell and its type system, as well as Propositions as types


I like to teach and learn: I have multiple teaching experience and I enjoy learning, be it on my research topics or something else


I am able to blitz a good meal usually vegetarian; I have a strong preference for mediterranean cuisine, but I also like to cook meals inspired by cambodian and japanese cuisine as well

Role-playing and tabletop game

I like to write tabletop RPG stories, and lose my mind watching players inventing completly different stories



Teaching assistant

IUT Orsay

Feb 2020 – May 2020 France
Teacher assistant for design and implementation of human-machine interface, for a total of 28 hours worked (part of it remotely due to COVID-19 pandemic).

Teaching assistant

Telecom Paris

Oct 2019 – Jan 2020 France
Teacher assistant for deep learning course for master students. 12 hours worked.

Teaching assistant


Nov 2018 – Dec 2019 France
Teacher assistant for C programming course for bachelor students, for a total of 30 hours worked.

PhD student


Oct 2018 – Present Saclay, France
Conducting research on formal methods applied to deep learning programs for verification

Recent Talks

Computational techniques for boosting verification of deep learning algorithms

A talk I gave in my lab, mostly to help me to structure and review my state-of-the-art

Formal verification of robustness properties in deep learning programs

A seminar I gave to people from Telecom, aimed at an audience familiar with machine learning but not familiar with verification …

Recent Publications

Partitionnement en régions linéaires pour la vérification formelle de réseaux de neurones

La grande polyvalence et les résultats impressionnants des réseaux de neurones modernes viennent en partie de leur non-linéarité. Cette …

CAMUS: A Framework to Build Formal Specifications for Deep Perception Systems Using Simulators

The topic of provable deep neural network robustness has raised considerable interest in recent years. Most research has focused on …


PFIA 2020

Theory and practice of deep learning verification; a tutorial I gave during the PFIA 2020 conference

ForMaL 2019

Some tutorial materials I gave with my advisor Guillaume Charpiat during DigiCOSME Spring school ForMaL