University of Edinburgh
Tuesday, February 24, 2015 at 2:30pm in Room 10 (Morgagni)
Analysis of large-scale social dynamics has historically been dominated by qualitative approaches relying on expert intuition and detailed but limited data (intelligence). Recent technological advances such as social networks are however increasingly producing large amounts of data which may enable a more quantitative analysis approach. Here, I will describe recent work where we used ideas from signal processing and Bayesian statistics to capture the dynamics of the Afghan conflict from the spatio-temporal pattern of event logs contained in the Wikileaks Afghan War Diary release. We show that the quantitative approach affords non-trivial insights in the conflict dynamics, and enables surprisingly accurate out of sample predictions. The talk is aimed at non-statisticians and I will (briefly) introduce the relevant concepts in point processes, spatio-temporal dynamical systems and Bayesian inference.
- Zammit-Mangion, Dewar, Kadirkamanathan and Sanguinetti, Point Process Modelling of the Afghan War Diary, PNAS 109 (31), 2012
- Zammit-Mangion et al, Modelling conflict dynamics with spatio-temporal data, SpringerBriefs in Applied Sciences and Technology, 2013
- Cseke, Zammit-Mangion, Heskes and Sanguinetti, Sparse approximate inference for spatio-temporal point process models, (online)
Guido Sanguinetti is a Reader in the School of Informatics at the University of Edinburgh. He holds a Laurea in Fisica from the Università di Genova and a DPhil in Mathematics from the University of Oxford. His research interests focus on Bayesian machine learning and applications, primarily to systems biology. He holds an ERC Starting Grant and was awarded the 2012 PNAS Cozzarelli Prize for Applied Science and Engineering.