Bayesian Methods for Animal Social Network Analysis

Modern technology allows researchers to use data loggers to automatically collect data on animal movements. This technology has the potential to detect behaviours of interacting animal but current machine learning methods for translating logger data into behaviours in complex social contexts are in their infancy and need further development. This project will develop machine learning methods for behavioural detection, and will use these advances to study animal social decision making.

Individuals show tremendous variation in their social behaviours and relationships, with consequences for their health and survival. There has been a surge of interest in the mechanisms that underpin this variation, with special attention paid to the decisions animals make navigating their social world. However, the majority of research to date treats individuals as actors in a social vacuum. But most societies are complex and social influences can override decisions made in isolation. We therefore need research aimed at understanding decisions within a broader social context. The main challenge is to document the social environment in sufficient detail; Human observations of simultaneous interactions are near impossible and technological innovations are needed. This project will use remote biologging technology and reality-mining to determine how and to what extent the social environment impacts upon individual decisions.

The broad aims of this project are two-fold. First, we will develop an automated system to detect social interactions using biologgers. Biologgers record data on acceleration, direction of movement, and distance to others. Biologgers will be deployed on individuals interacting in a social context and machine-learning algorithms will be used to automatically classify interactions, which will be validated with manual observations. Humans and domestic animals (e.g. dogs) will be potential study subjects. Second, these unique "big" social data, will be used to determine how and to what extent the social environment impacts individual decisions. This project represents the first attempt to record active social interactions using remote data and we expect both its technological advances and empirical results will be highly influential in disciplines from computer science to psychology.