Quantifying social roles in multi-animal videos using subject-aware deep-learning.

Area of Science:

• Neuroscience
• Psychology
• Ecology
• Ethology
• Animal Behavior Analysis

Background:

• Analyzing social behavior is crucial across multiple scientific disciplines.
• Existing computational tools are limited to analyzing animals in controlled settings with restricted social interactions.
• Automated identification of social roles in freely moving, multi-animal groups remains a significant challenge.

Purpose of the Study:

• To develop and present a novel deep-learning-based system, LabGym2, for the automated identification and quantification of social roles in multi-animal groups.
• To introduce a subject-aware approach that considers individual behavior within its social and environmental context.

Main Methods:

• Developed LabGym2, a deep learning system employing a subject-aware methodology.
• Evaluated the system's performance across various species (insects, primates) and experimental setups (partner preference assays, field observations).

Main Results:

• Demonstrated the efficacy of the subject-aware deep learning approach in identifying and quantifying social roles in diverse animal groups.
• Successfully applied the system to analyze insect partner preference and primate social interactions in naturalistic settings.

Conclusions:

• The subject-aware deep learning framework offers a controllable, interpretable, and efficient method for analyzing complex social dynamics.
• This approach facilitates new experimental designs and systematic evaluations of interactive behaviors within animal groups.