Abstract
Optimal control of diffusion processes is intimately connected to the problem of solving certain Hamilton–Jacobi–Bellman equations. Building on recent machine learning inspired approaches towards high-dimensional PDEs, we investigate the potential of iterative diffusion optimisation techniques, in particular considering applications in importance sampling and rare event simulation, and focusing on problems without diffusion control, with linearly controlled drift and running costs that depend quadratically on the control. More generally, our methods apply to nonlinear parabolic PDEs with a certain shift invariance. The choice of an appropriate loss function being a central element in the algorithmic design, we develop a principled framework based on divergences between path measures, encompassing various existing methods. Motivated by connections to forward-backward SDEs, we propose and study the novel log-variance divergence, showing favourable properties of corresponding Monte Carlo estimators. The promise of the developed approach is exemplified by a range of high-dimensional and metastable numerical examples.
Highlights
Hamilton–Jacobi–Bellman partial differential equations (HJB-PDEs) are of central importance in applied mathematics
Building on recent machine learning inspired approaches towards high-dimensional PDEs, we investigate the potential of iterative diffusion optimisation techniques, in particular considering applications in importance sampling and rare event simulation, and focusing on problems without diffusion control, with linearly controlled drift and running costs that depend quadratically on the control
We show that a variety of loss functions can be constructed and analysed in terms of divergences between probability measures on the path space associated to solutions of (1), providing a unifying framework for iterative diffusion optimisation (IDO) and extending on previous works in that direction [59,73,128]
Summary
Hamilton–Jacobi–Bellman partial differential equations (HJB-PDEs) are of central importance in applied mathematics. We show that a variety of loss functions can be constructed and analysed in terms of divergences between probability measures on the path space associated to solutions of (1), providing a unifying framework for IDO and extending on previous works in that direction [59,73,128]. As this perspective entails the approximation of a target probability measure as a core element, our approach exposes connections to the theory of variational inference [15,124]. The aforementioned adjustments needed to establish the path space perspective often lead to faster convergence and more accurate approximation of the optimal control, as we show by means of numerical experiments
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have