Séminaire IMAGE : « Few-Shot Bioacoustic Sound Event Detection and Classification », Ilyass Moummad

Nous aurons le plaisir d’écouter Ilass Moummad, doctorant de l’IMT Atlantique, Brest.
Il donnera un séminaire IMAGE  le jeudi 2 octobre à 14h00 en salle de séminaire F-200.

Titre : « Few-Shot Bioacoustic Sound Event Detection and Classification »

Résumé :
Bioacoustics is a cross-disciplinary field combining biology and acoustics to study how animals produce, transmit, and receive sound. This research provides key insights into animal behavior, communication, and environmental interactions. Deep learning has become a powerful tool for decoding complex animal vocalizations, aiding biodiversity monitoring and conservation efforts. However, collecting large annotated datasets remains challenging, particularly for rare species or remote areas. To address this, few-shot learning offers a solution by training models to detect and classify animal sounds with minimal labeled data. Additionally, self-supervised learning can leverage easily collectible, unlabeled animal sound data to learn representations without the need for extensive annotations. We address these challenges by utilizing recent advancements in representation learning, particularly contrastive learning. Our experiments on public benchmarks demonstrate the effectiveness of contrastive learning in improving generalization for bioacoustic monitoring.

Bio :
Ilyass Moummad is a PhD student (December 2021 – November 2024) at IMT Atlantique, Brest, France. He works under the supervision of Nicolas Farrugia (IMT Atlantique) and is co-supervised by Romain Serizel (Inria Multispeech). Ilyass’s PhD topic is Deep Learning for Bioacoustics, with an interest in representation learning (both self-supervised and supervised) of animal sounds, as well as few-shot learning (species sound classification and detection from very few annotated examples).