Deep generative modelling of behaviour

Lead Research Organisation: University College London
Department Name: Institute of Health Informatics

Abstract

The objective of this proposal is to construct a high-fidelity generative model of human behaviour in a clinical setting, derived from the analysis of anatomically labelled patient movement. A generative model of behaviour is a statistical model that comprehensively describes the distribution of its possible forms. Such a model would be invaluable in enabling the detection of behaviours too rare to be learnt by conventional statistical models that merely discriminate one form of behaviour from another. Since the space of possible movements is very large, no satisfactory generative model currently exists. The novel engineering content of this proposal is the design, estimation and evaluation of a high-fidelity model based on a new deep artificial neural network architecture specifically tailored to the task. The student will take the approach of using large-scale anonymized data drawn from a hospital patient population in combination with a set of novel deep neural network architectures to create an array of candidate models. The optimal model will be selected by its performance in detecting anomalous behaviours, tested on unseen, out-of-sample data. The task will require joint innovation in the fundamentals of complex generative modelling of high-dimensional time series data and its application to healthcare in the context of a diversity of abnormal behaviours.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S021612/1 01/04/2019 30/09/2027
2248041 Studentship EP/S021612/1 01/10/2019 30/09/2023 Anthony Peter Bourached