Jean-Sébastien Giroux

MASc Student · Polytechnique & Mila, Montréal, Québec

I am a master's student working at the intersection of artificial intelligence and neuroscience, a field known as Neuro-AI. My work spans two complementary areas: Artificial Intelligence, where I use mathematical optimization - Bayesian Optimization based on Gaussian Processes (GPBO), and Neuroscience, where I use neuromodulation to restore lost functions in animals and humans after an injury. I am interested in fundamental AI, mathematics research, neuroscience and in bringing them to operational use in the medical field. My research is done under the supervision of Professor Marco Bonizzato @ Polytechnique Montréal and Mila - Québec AI institute within sciNeurotech Lab.

Before joining this lab, I was an applied research scientist @ Environment and Climate Change Canada leading two projects - Wind power downscaling over subcontinental grid, and wind power forecasting for Quebec wind farms.

I am very interested in bringing my research to operational use - don't hesitate to contact me for collaborations.

Affiliations

  • Polytechnique Montréal & Mila – Québec AI Institute
    Master Student · Supervisor: Prof. Marco Bonizzato     2026 — Present
  • Startup (stealth mode)
    CEO · Venture Scientist     2026 — Present
  • Environment and Climate Change Canada (ECCC)
    Applied AI Research Scientist     2023 — 2026
  • Sherbrooke University – GRAMS Lab
    Research Intern · Medical Imaging & Deep Learning     January - April 2023

Research

Interests

Artificial Intelligence Deep Learning Computer Vision Time Series Multimodal Models Foundation Models Neuromodulation Neuroscience
Meteorological-AI Ongoing

Deep Learning Wind Power Forecasting with Multimodal Sequence Models

Jean-Sébastien Giroux*, Thomas Forest*, Mathieu Prévost et al. · Work in Progress

Accurate short-term wind power forecasting is essential for grid stability and renewable energy integration. This project provides a comprehensive benchmark of deep learning architectures - Dense networks, Recurent neural networks (LSTMs, GRUs), Transformers, and the recent Mamba state-space model - for next-day wind power forecasting at 1-hour temporal resolution over wind farm in Québec.

We combine two complementary data sources: SCADA data (real-time sensor readings from the wind park) and NWP data (numerical weather prediction forecasts). Fusing both sources consistently outperforms models trained on either alone, establishing a strong benchmark for operational wind power forecasting.

📄 Paper (coming soon) 💻 Code (coming soon)
Meteorological-AI Published · Artificial Intelligence for the Earth Systems (AIES) 2025

Interpolation-Free Deep Learning for Meteorological Downscaling on Unaligned Grids across Multiple Domains with Application to Wind Power

Jean-Sébastien Giroux, Simon-Philippe Breton, and Julie Carreau. · AIES, 2025

As wind energy production accelerates, reliable wind forecasts become critical — but numerical weather prediction models are too computationally expensive to capture fine-scale wind behaviors. We develop a deep learning downscaling model based on a U-Net architecture that learns to map coarse probabilistic wind forecasts to high-resolution outputs, eliminating the need for conventional preprocessing steps. The model is extended to the full Canadian region via transfer learning, and results show that low-resolution wind speed alone is sufficient as input. Downscaled wind velocities show promising improvements in detecting wind power ramps - a key phenomenon for grid stability.

📄 Paper 💻 Code