AI4sustainability is a new research project that we are currently launching at Télécom Paris in 2023-24.
WANTED ! We are currently searching for industrial and institutional sponsors !
New: with Charlotte Delage, Charlotte Laclau, Rémi Flamary and Karim Lounici we are organizing the Ellis Doctoral Symposium on AI & Sustainability in Paris from Aug 26, to AUg 29, 2024.
SUSTAINABLE AI
The first component of the project is devoted to invent and develop novel models and algorithms to ensure a sustainable AI aka trustworthy and frugal AI.
Frugality in AI/Machine Learning
- Low-rank approaches, including sketching (see PhD thesis of tamim El Ahmad, and Luc Brogat-Motte): efficient approaches to Structured prediction
- Data-efficient methods: active learning, few-shot learning
- Model-Efficient methods: regularization, chemistry-aware approaches, multi-tasks, infimum of tasks learning
- Objective-driven approaches
Trustworthy AI
- Robust distances and metrics
- Learning with robust losses
- Mixup methods
- Learning with interpretability (see our papers @NeurIPS 21 and 22, Tr. Audio, Speech, Lang. processsing 2024, ICLR 2025)
- Post-hoc interpretability
- By-design interpretability
AI FOR SUSTAINABILITY
The second component of the project consists in leveraging AI tools to help to address some of the issues related to sustainable goals as defined by UN.
AI for monitoring and forecasting key resources:
- Renewable energy: forecasting of energy production in wind turbine
- Load forecast of electric vehicle charging station and beyond: IP Paris (E4C) PhD thesis of Wen Yang: Resource availability forecasting, co-supervised with Mathieu Fontaine
AI for biodiversity surveillance
- Analyzing Fish resources
- Measuring biodiversity
AI for health & living systems – in progress
=== Current support===
ANR project (PEPR IA) : FOUNDRY