Seminar: On-line inference for emerging networks and group structures

SpeakerSimon Godsill
AffiliationCambridge University Engineering Department
DateThursday, 31 May 2012
Time14:00 - 15:00
LocationMalet Place Engineering Building 6.12
Event seriesStatistical Science Seminars

In this talk I will describe Bayesian methods for on-line learning about the structure of a scene involving multiple interacting agents. The models we use are interacting continuous-time stochastic differential equations, which are able to capture the attractive forces that describe the motion of individuals who are attempting to follow some kind of `group' structure, for example convoys of vehicles or aircraft. Sequential Monte Carlo methods are used to sample the state variables in the system, which include a group network structure that can vary over time (e.g. a new group forms, a group splits, a new object appears/disappears, etc.). The models are tested on tracking scenarios and have also proved useful for financial modleling applications where certain commodities follow similar patterns over time.

iCalendar csml_id_99.ics