• Title: “Self-supervised learning on heterogeneous graphs: Contribution to the analysis of animal behaviors and social dynamics” ; INSA.
  • Postdoctoral researcher: Vincent Elouan.
  • Affiliated unit: INSA, DM2L.
  • Supervision:
    • Céline Robardet (INSA).
  • Project duration: 2024–2027.

Project summary:

Methodological advances are necessary in the study of animal welfare to better understand animal interactions and their behaviors. Traditional machine learning approaches on graphs often involve embedding these discrete structures into Euclidean spaces, leveraging annotations to structure the space. However, these annotations can be challenging to obtain.
Objectives:

• Address these challenges through self-supervised learning paradigms while improving model result interoperability.

Methodologies:

• Analyze heterogeneous graphs that capture complex animal behavior patterns.