Deep learning of activity monitoring data for disease detection to support livestock farming in resource-poor communities in Africa

The project:
The aim of this 3-year PhD studentship project is to employ state-of-the-art machine learning methods to support sustainable livestock health and production in the poorest rural communities in Africa. The project’s partners have successfully trialled a low-cost autonomous system of animal-worn accelerometers that communicate with a solar-powered control station which relays the data to a central server over the mobile network (http://hdl.handle.net/2263/53304). However, a high degree of daily variation tends to hide changes in animal health and management issues, precluding the use of simple data analysis approaches.

Our data necessitates consideration of individual-level behaviour as deviations from herd-level behaviours. To achieve this level of analysis, state-of-the-art machine learning approaches will be investigated and developed. We have many months of data on hundreds of animals across a number of farms, with on-going data collection continually expanding. As an example, a multivariate Long Short-Term Memory (LSTM) autoencoder could be trained for the unsupervised modelling of our extensive data on ‘normal’ behaviour in order to detect anomalies both at short (e.g. animal theft) and long (e.g. the effects of chronic disease) timescales. Bristol’s GPU-powered BlueCrystal4 supercomputer (https://www.top500.org/site/49136) is readily available to support computational efforts. Further into the project, we will take advantage of the continuous stream of new data to refine our models using currently evolving online deep learning techniques, and may also explore deep reinforcement learning for explicit model discovery where only long-term quality parameters are available.

This three year studentship can start around September 2018 or before. The student will be based within Prof Dowsey’s Data Science group in the Department of Population Health Sciences, Bristol, and will benefit from an interdisciplinary supervision team from the Department of Computer Science (Dr Tilo Burghardt, with expertise in machine learning for animal biometrics; e.g. https://goo.gl/v7LPb1) and the School of Biological Sciences (Dr Christos Ioannou, an expert in animal behaviour). The student will benefit from membership of the Bristol Doctoral College, which provides a comprehensive support and training programme for PhD candidates, and will join an established international consortium of PhD students and researchers all working on sustainable food security in southern African).

Candidate requirements: The studentship would suit an applicant with a strong first degree or masters in a computational discipline (e.g. mathematics, computing, electrical engineering) and competent programming skills (preferably knowledge of Python, C++ and Matlab). A good understanding of the Tensorflow framework is an advantage.

How to apply:
Please make an online application for this project at http://www.bris.ac.uk/pg-howtoapply. Please select ‘Faculty of Health Sciences’ and then ‘Veterinary Sciences_(PhD)’ on the Programme Choice page and enter details of the studentship when prompted in the Funding and Research Details sections of the form.

Contacts: Prof Andrew Dowsey ([Email Address Removed])