Léonard Boussioux

Multimodal Artificial Intelligence for Wildlife Analytics

Léonard Boussioux, Charles Kantor, et al.

Abstract

While worldwide ecosystems face a mass extinction of species, demographic data related to shifts in species diversity and abundance has substantial taxonomic, spatial, and temporal biases and gaps. Available methods to study and monitor species and their population trends are often antiquated and inefficient. There is a need for efficient, rigorous, and reliable methods to study and monitor wildlife. We introduce a systematic and holistic framework to build efficient AI tools adapted to wildlife data, challenges, and needs. We illustrate our methodologies with real-world datasets provided by several museums and crowdsourcing platforms and show the impact of our state-of-the-art models.

Over-MAP: Structural Attention Mechanism and Automated Semantic Segmentation Ensembled for Uncertainty Prediction

C. Kantor, L. Boussioux, B. Rauby, H. Talbot

Proceedings of the AAAI Conference on Artificial Intelligence

Asymptotic cross-entropy weighting and guided-loss in supervised hierarchical setting using deep attention network

C. Kantor, L. Boussioux, E. Jehanno, H. Talbot

AAAI Fall Symposium on AI for Social Good

Gradient-Based Localization and Spatial Attention for Confidence Measure in Fine-Grained Recognition using Deep Neural Networks

C. Kantor, L. Boussioux, B. Rauby, H. Talbot

Proceedings of the AAAI Conference on Artificial Intelligence

Geo-Spatiotemporal Features and Shape-Based Prior Knowledge for Fine-grained Imbalanced Data Classification

C. Kantor, M. Skreta, B. Rauby, L. Boussioux, E. Jehanno, A. Luccioni, D. Rolnick

IJCAI 2020