Individualisation of Behaviour Models and Identifying Social Hierarchies
The analysis of animal (and human) behaviour has found applications in many fields (conservation, demographics, medicine etc...). However, most behavioural models in literature either deal with absolute positions only and infer a few coarse high-level behavioural states, or treat fine-grained activities in terms of frequencies of manifestation with little to no temporal modelling. At the same time, the analysis has to contend with noisy observations, such as errors in human annotations, which are often ignored in the interest of simplicity.
In this project we will attempt to address some of these issues in the context of mouse-behaviour phenotyping, in collaboration with the MRC Harwell Institute. The main goal of the biologists in this case is to analyse and differentiate between various knockout and wild-type genetic strains. We propose building Hierarchical behavioural models to better capture the nuances which may manifest themselves when it comes to behaviour differences and which may not be readily apparent from summary statistics of the data. We aim to do this analysis using various data modalities, including the annotated behaviour labels, automatically collected position data as well as infra-red video.