advanced learning technologies
In a fast-moving digital world, driven by Data and Machine Learning techniques, our competence center uses artificial intelligence to solve complex industrial problems. We accompany the need for novel and fitted AI-based architectures taking into account operational locks to roll out these techniques.
Given the market uncertainty and agility of the major digital operators, a lot of industrialists are forced to bypass their usual patterns and invest in less mature technologies to remain competitive.
Our objective is to gather common investments in AI and capitalize on R&T in a competitive sector undergoing deep restructuring.
As we are fully aware of the need to provide reliable and explainable AI, we tackle problems in which data frugality, embedding AI, Man-Machine interactions, and decision-making raise challenges.
Born to serve spatial applications, our competence center keeps widening its application sectors to fully reach the aerospace and transportation industries, as well as the health sector.
FRUGAL AI AND LIFECYCLE MANAGEMENT
Our research focused on several themes that share a common challenge: the lack of training data:
- Unsupervised and generative techniques: Domain Adaptation, Representation Learning;
- Data augmentation, synthetic datasets usage;
- Incremental learning and Graph Neural Nets representations;
- Development process (design pattern) and incremental maintenance/correction of AI-based systems;
- Embedded AI (infrastructure and hardware architectures).
OPTIMIZATION AND DECISION-MAKING
Beyond traditional optimization methods, the latest AI progress recently demonstrated relevance in learning decision strategies in complex environments, highly combinatory, and subject to uncertainties.
In this context, we develop AI technologies, proof-of-concept, and demonstrators based on the following approaches:
- Multi-agent architectures;
- Reinforcement learning techniques;
- Genetic Programming.
The next generation of critical systems (from factory automation to the single-pilot operation of airplanes) will host AI components interacting with Humans.
Here, our experts articulate their research around one main goal, which is to pave the way to bring together AI and humans into the loop with:
- Explainability and interpretability of AI outcomes;
- Optimization of Human-Machine teaming and monitoring;
- Learning with and from Human: Deep Learning techniques use massively automated algorithms that need millions of runs. Integrating Human into the learning process requires a rethinking of machine learning techniques.
2019 > 2025
Contact our platform experts team