Richard Everitt, University of Oxford
Anthony Lee, Warwick University
Nicolas Kantas, Imperial College London
Nick Whitely, Bristol University
Simon Maskell, QinetiQ & Imperial College London
The explosion of research in Monte Carlo methodology during the past 20 years has resulted in many application domains being analysed that were hitherto deemed impossible. It is the aim of this meeting to deliver an overview of recent advances in Monte Carlo methodology, the problems that remain to be solved, and some applications.
Richard Everitt: Missing data, and what to do about it. This talk describes two approaches to inference in the presence of missing or incomplete data: approximate Bayesian computation (ABC) and particle MCMC. Both methods are applied to parameter estimation of a hidden Markov random field, and are compared to the standard data augmentation approach.
Anthony Lee: Auxiliary variables and many-core computation. The use of auxiliary variables in various Monte Carlo methods has proliferated both explicitly and implicitly over the last two decades, as our understanding of how to devise effective algorithms has grown. In addition, massively parallel 'many-core' processors have become the focus of the high performance computing community for a variety of physical reasons. The confluence of these two streams of research provides an opportunity for novel development and re-assessment of algorithms using this parallel model of computation. I will overview some promising strategies, new developments and open questions in this area of research.
Nicolas Kantas: Particle methods for computing optimal control inputs. In this talk we consider how particle methods can be used for risk sensitive control problems. The approach is
based on earlier efforts for deterministic systems and is essentially a sequential Monte Carlo
(SMC) algorithm for an appropriate dual ﬁltering problem. We will describe how the approach can be used for scheduling unit commitment for power systems.
Nick Whitely: Stability properties of some particle filters. Whilst there is a rich literature on various theoretical properties of particle filters, most existing results which explain their stability properties rely on strong mixing assumptions which do not hold in typical applications. One of the main obstacles is dealing with a non-compact state space. This talk will describe recent developments which allow some stability properties of particle filters to be verified when the state space is non-compact.
Simon Maskell: Using a Probabilistic Hypothesis Density filter to confirm tracks in a multi-target environment. We describe a real-time scalable multi-target particle filter tracking, applicable to
tracking hundreds of targets, each of which is modelled with a low-dimensional continuous state (eg position and velocity in three dimensions). The approach is shown to maintain good tracking performance for high clutter, low-detection scenarios where Kalman filter based multi-target trackers fail to perform well. We also show results from a scenario with obscured regions where the target cannot be detected, and show that targets can be tracked through the obscurations.
Meeting organised by General Applications Section
Meeting organiser: Mark Briers
Contact details: email@example.com
Please download a booking form www.rss.org.uk/eventforms