Bridget Smart
PhD student at the University of Oxford
Talk recording
When we are looking to predict sequences of interactions, movements or events in complex systems, it is often necessary to account for temporal memory and structural correlations in our models. Traditional higher-order Markov models are often applied to capture such dependencies, but at the cost of exponential state-space growth and limited interpretability. In this talk we will introduce the Separable Markov model, a subset of higher-order Markov models which assume a linearised dependence on the history of the process. This assumption significantly reduces the number of parameters in the model, making the Separable Markov model a promising alternative in settings with limited data or long-range memory. We discuss the properties of this model, how its expressivity and interpretability can be evaluated and show examples of performance across a range of real-world datasets.
About the speaker
Bridget is a Rhodes Scholar pursuing a DPhil at the University of Oxford’s Mathematical Institute and Institute of New Economic Thinking, supervised by Professor Renaud Lambiotte and Professor J. Doyne Farmer. Her research is focused on developing theoretical tools grounded in information theory, network science and probability to model, characterise and assess uncertainty in complex real-world systems. Her current work explores models for dynamics with long-range memory on networks, prediction in systems with complex interactions and emergent effects, and the dynamics of social systems.
Share this page:
