arXiv subjects

Tony Shardlow

Publications and source records attributed to Tony Shardlow.

21 records · Page 2Linked to original sources

arXiv2017

Langevin equations for landmark image registration with uncertainty

Registration of images parameterised by landmarks provides a useful method of describing shape variations by computing the minimum-energy time-dependent deformation field that flows one landmark set to the other. This is sometimes known as the geodesic interpolating spline and can be solved via a Hamiltonian boundary-value problem to give a diffeomorphic registration between images. However, small changes in the positions of the landmarks can produce large changes in the resulting diffeomorphism. We formulate a Langevin equation for looking at small random perturbations of this registration. The Langevin equation and three computationally convenient approximations are introduced and used as prior distributions. A Bayesian framework is then used to compute a posterior distribution for the registration, and also to formulate an average of multiple sets of landmarks.

math.NA↗

arXiv2015

On the pathwise approximation of stochastic differential equations

We consider one-step methods for integrating stochastic differential equations and prove pathwise convergence using ideas from rough path theory. In contrast to alternative theories of pathwise convergence, no knowledge is required of convergence in pth mean and the analysis starts from a pathwise bound on the sum of the truncation errors. We show how the theory is applied to the Euler-Maruyama method with fixed and adaptive time-stepping strategies. The assumption on the truncation errors suggests an error-control strategy and we implement this as an adaptive time-stepping Euler-Maruyama method using bounded diffusions. We prove the adaptive method converges and show some computational experiments.

math.NA↗

arXiv2014

Improving MLMC for SDEs with application to the Langevin equation

This paper applies several well-known tricks from the numerical treatment of deterministic differential equations to improve the efficiency of the Multilevel Monte Carlo (MLMC) method for stochastic differential equations (SDEs) and especially the Langevin equation. We use modified equations analysis to circumvent the need for a strong-approximation theory for the integrator, and we apply this to introduce MLMC for Langevin-type equations with integrators based on operator splitting. We combine this with extrapolation and investigate the use of discrete random variables in place of the Gaussian increments, which is a well-known technique for the weak approximation of SDEs. We show that, for small-noise problems, discrete random variables can lead to an increase in efficiency of almost two orders of magnitude for practical levels of accuracy.

math.NA↗