Citizen science & machine learning for plant Identification: challenges from Pl@ntNet

Orateur:
Joseph Salmon
Localisation:
Type: Séminaire de probabilités et statistiques
Site: UGE , 4B 125
Date de début:
Date de fin:

Pl@ntNet is a global citizen science platform that harnesses artificial intelligence to enable
large-scale plant species identification and biodiversity monitoring. With 10 millions of contributors
across over 200 countries, it represents one of the world’s largest biodiversity observatories. While
such platforms offer a cost-effective and scalable approach to data collection, the quality of user-
generated annotations can vary significantly. This variability can affect the reliability of automated
plant identification, which is key for both research and real-world biodiversity monitoring
applications.
In this talk, I will present how to leverage statistically grounded prediction methods, including conformal prediction, to provide valid confidence sets for species identification, specifically optimized for long-tail classification scenarios.
This is based on joint work with Tiffany Ding (UC Berkeley), Jean-Baptiste Fermanian (Inrae) and the Pl@ntNet team.