Deep Learning in Dynamic Systems
Social systems distinguish themselves from most physical systems in that agents of the system constantly change the rules they play by. Planetary motions, for instance, follow the same rules throughout space and time, thereby allowing highly accurate predictions. In contrast, patterns of human behavior are in constant flux. This makes it extremely hard to extrapolate known patterns in to the future. Traditional Machine Learning algorithms are ill equipped to face this challenge, as they are designed to learn once how to generate predictions and so fail to update its representation of the system as the dynamics change.
This project aims to leverage Deep Learning and stochastic modelling to build scalable predictive models that dynamically update internal hypotheses about the likely trajectory of a dynamic system. The project specifically aims to overcome difficulties in overfitting trends in historic data while simultaneously allowing early detection of true trend changes.
April 2017 – April 2020
This project is funded by the ESRC.
Sebastian joined the Ph.D. program in April 2017. He previously worked as a Data Scientist at the Boston Consulting Group (BCG Gamma), where he focused on leveraging machine learning for predictive modeling. Sebastian holds an M.Sc. in Economics from the Stockholm School of Economics.
- Prof Mark Elliot (School of Social Statistics)
- Prof John Keane (Department of Computer Science)
Office: G45, Humanities Bridgeford St Building