↓ Skip to main content
SubAlpine.IT
SubAlpine.IT

Who Am I?

Jérôme Buisine
Hi! I'm Jérôme Buisine, PhD in Computer Science and founder of SubAlpine.IT. I'm sensitive to current ecological and climate issues, and my job is to adjust and conceptualise machine learning models in order to meet specific needs — no more, no less.

Background and experiences

  1. Freelance Data Scientist

    2024 - Now.
    @SubAlpine.IT - France
    • Machine learning & AI solutions
    • Optimisation (metaheuristics, operational research)
    • Data science & statistical analysis
    • Software development (Python backend, PostgreSQL, mobile & web)
    • Scientific consulting & R&D

  2. Associate Professor in IT

    2022 - 2023
    @LISIC - Calais (France)
    • Machine learning methods to optimise time and cost in synthetic imaging
    • Computer graphics research
    • University teaching (computer science, data science)

  3. Temporary Lecturer and Research Assistant

    2021 - 2022
    @LISIC - Calais (France)
    • Research on noise perception in computer-generated images
    • Teaching programming and machine learning

  4. PhD in IT

    2018 - 2021
    @LISIC - Calais (France)
    Subject: Machine learning methods for taking into consideration noise in computer generated images.

    Carried out within the ANR PrISE-3D project (Perception, Interaction and 3D Lighting Simulation).

  5. MSc in IT (with distinction)

    2016 - 2018
    @ULCO - Calais (France)

Research

The thread running through my research: teaching machines to perceive the noise in a rendered image as finely as the human eye — and to recognise the moment an image is “clean enough” to stop rendering. The payoff: compute time and energy saved.

This work grew around a few axes that feed into one another:

  • Image synthesis — Monte-Carlo rendering, noise and artifact (firefly) removal, and automatic stopping criteria for rendering.
  • Statistical characterisation & feature selection — describing noise through statistical and information-theoretic measures (Kullback-Leibler divergence, Hellinger distance, statistical moments, entropy…), then isolating the most discriminative features for the models.
  • Deep learning — neural architectures to characterise and detect noise where classical measures reach their limits.
  • Visual perception — eye fixations and perception thresholds, to calibrate the models on what the eye actually sees rather than on an arbitrary metric.

This work has led to publications in journals and conferences such as Journal of Vision, Entropy, ICMLA, the ACM Symposium on Applied Perception and the Eurographics Symposium on Rendering, as well as an open-source optimisation library published in the Journal of Open Source Software.

You can find my publications on Google Scholar and ResearchGate, my code on GitHub, and more details on my personal website: jeromebuisine.fr.