Laure Bourgois
Engineer-trainer specialized in AI, AI Ambassador, Associate Professor at the University of Paris-Saclay
Doctor in artificial intelligence, Laure Bourgois is an "AI Ambassador" (national plan "Dare AI"). With over 24 years of experience in R&D (Orange Labs, IFSTTAR ...), Laure Bourgois is also commissioned by training organizations and universities on Deeptech topics. Between 2018 and 2023, she was co-founder and president of Codataschool, a professional training organization specialized in AI and Big Data. She joined Inria (National Institute for Research in Digital Science and Technology) as an engineer on Open Source projects in June 2023. She is also an associate professor at the University of Versailles in the master's program in AI since July 1, 2024. She has published about ten scientific and popular articles on AI. Her expertise covers the following topics around AI: - Machine learning, - Multi-agent simulations, with applications in simulating pedestrian and crowd behavior, - Semantic modeling of data and services. As a committed computer scientist, she is a member of the Frederik Bull Institute and voluntarily conducts AI workshops for the @Maryse Project association (which works for diversity in tech in overseas territories). She has also been a sponsor of the Women's Mathematics Olympiad for the Stem4all association.
Prices
- Conference : 2200 €
Localization
Languages
My conferences
WILL AI BE ABLE TO ANTICIPATE OUR BEHAVIORS? Interview for 697IA with Jérôme Bonaldi
Laure Bourgois is interviewed by the media 697IA by Jérôme Bonaldi about crowd simulation. She presents the scientific and operational work to model and simulate pedestrian behavior. 1. Applications : These simulations have multiple applications : - Urban planning and design (development of public spaces), - Safety and security (management of evacuations, prevention of stampedes), - Commercial use (placement of advertising panels), - Video games and cinema (populating scenes with realistic avatars). She cites notable examples: the Love Parade stampede in Germany (2010) or the pilgrimages to Mecca, where poorly managed crowd movements had dramatic consequences. 2. Evolution of models The first simulations, in the 1970s, treated pedestrians as fluids (equations of fluid mechanics). In 1987, Craig Reynolds simulated flocks of birds and schools of fish with simple behavioral rules. But for pedestrians, it is more complex because behaviors are not standardized. 3. The multi-agent approach and forces AI intervenes here through multi-agent systems: each pedestrian is an artificial agent immersed in an environment, with perceptions and decisions. The model used is that of repulsive and attractive forces: - Obstacles and other pedestrians exert a repulsive force. - Goals (following someone, fleeing a danger) create attractive forces. This model allows simulating crowd reactions to a danger (fire, attack) and anticipating collective movements. 4. Challenges and specificities - Computing power: simulations are costly, but they can be lightened by reducing the perception range of each agent (pedestrians only interact with their immediate surroundings). - Cultural behaviors: models must be calibrated according to countries. For example, the French cross outside of crosswalks, which is not the case in Switzerland or Germany. Behavioral psychology researchers have helped integrate these cultural biases. - Hybridization of models: depending on the situation, one can use a fluid model (for dense flows) or a particle model (for individual movements). The challenge is to switch from one to the other without losing information. 5. Final objective : Save lives, optimize evacuations, and better understand human behaviors in stressful situations or dense crowds.