Search site

School of Geography

Propagation of Error and Emergent Properties in Agent-based simulations

Supervisors: Dr Andy Evans and Dr Alison Heppenstall

Agent-based models (think "the Sims"; "SimCity") are an increasingly used modelling architecture for social and environmental models. They are relatively simple to build, can integrate behavioural and statistical information, and their form matches human perceptions of the real world. Moreover, agent-based models fit well with current paradigms of emergence and complexity - the notion that simple rules at the level of individuals in a system can generate either simple or complex patterns when grouped together and allowed to interact.

However, such models have problems. The very complexity of the interactions between multiple individuals in a system (eg. economic models, or models of ecosystems) means that it is hard to trace the development of emergent group-level patterns, and, even more so, say how reliable these patterns are. In geographical systems, geographical correlation (the notion that things near each other are similar / have an effect on each other) makes this doubly complicated.

This PhD will investigate and develop methodologies for tracing the propagation of error and the development of emergent properties through such complex agent-based systems. The methodologies will be part modelling, part data exploration, and part visualization. The methodologies will be general, though there will be the opportunity to apply the techniques to a wide variety of test-bed systems, from retail markets to disease spread, dependent on the student's interests.

This PhD would suit a computer science, physics or mathematics student, or a geography student with a strong interest in quantitative methods and modelling. The student would be associated with the Multi-Agent Systems and Simulation research group, which is part of the School's Centre for Spatial Analysis and Policy.

Funding

For information on funding opportunities click here