Source author record

Ian Langmore

Ian Langmore appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Computation math-ph math.MP Machine Learning math.AP math.ST Methodology physics.ao-ph physics.comp-ph Programming Languages Statistics Theory

Catalog footprint

What is connected

6works

11topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Hamiltonian Monte Carlo in Inverse Problems; Ill-Conditioning and Multi-Modality

The Hamiltonian Monte Carlo (HMC) method allows sampling from continuous densities. Favorable scaling with dimension has led to wide adoption of HMC by the statistics community. Modern auto-differentiating software should allow more widespread usage in Bayesian inverse problems. This paper analyzes two major difficulties encountered using HMC for inverse problems: poor conditioning and multi-modality. Novel results on preconditioning and replica exchange Monte Carlo parameter selection are presented in the context of spectroscopy. Recommendations are given for the number of integration steps as well as step size, preconditioner type and fitting, annealing form and schedule. These recommendations are analyzed rigorously in the Gaussian case, and shown to generalize in a fusion plasma reconstruction.

preprint2020arXiv

A Condition Number for Hamiltonian Monte Carlo

Hamiltonian Monte Carlo is a popular sampling technique for smooth target densities. The scale lengths of the target have long been known to influence integration error and sampling efficiency. However, quantitative measures intrinsic to the target have been lacking. In this paper, we restrict attention to the multivariate Gaussian and the leapfrog integrator, and obtain a condition number corresponding to sampling efficiency. This number, based on the spectral and Schatten norms, quantifies the number of leapfrog steps needed to efficiently sample. We demonstrate its utility by using this condition number to analyze HMC preconditioning techniques. We also find the condition number of large inverse Wishart matrices, from which we derive burn-in heuristics.

preprint2020arXiv

tfp.mcmc: Modern Markov Chain Monte Carlo Tools Built for Modern Hardware

Markov chain Monte Carlo (MCMC) is widely regarded as one of the most important algorithms of the 20th century. Its guarantees of asymptotic convergence, stability, and estimator-variance bounds using only unnormalized probability functions make it indispensable to probabilistic programming. In this paper, we introduce the TensorFlow Probability MCMC toolkit, and discuss some of the considerations that motivated its design.

preprint2011arXiv

A Hybrid (Monte-Carlo/Deterministic) Approach for Multi-Dimensional Radiation Transport

A novel hybrid Monte Carlo transport scheme is demonstrated in a scene with solar illumination, scattering and absorbing 2D atmosphere, a textured reflecting mountain, and a small detector located in the sky (mounted on a satellite or a airplane). It uses a deterministic approximation of an adjoint transport solution to reduce variance, computed quickly by ignoring atmospheric interactions. This allows significant variance and computational cost reductions when the atmospheric scattering and absorption coefficient are small. When combined with an atmospheric photon-redirection scheme, significant variance reduction (equivalently acceleration) is achieved in the presence of atmospheric interactions.

preprint2011arXiv

Importance Sampling and Adjoint Hybrid Methods in Monte Carlo Transport with Reflecting Boundaries

Adjoint methods form a class of importance sampling methods that are used to accelerate Monte Carlo (MC) simulations of transport equations. Ideally, adjoint methods allow for zero-variance MC estimators provided that the solution to an adjoint transport equation is known. Hybrid methods aim at (i) approximately solving the adjoint transport equation with a deterministic method; and (ii) use the solution to construct an unbiased MC sampling algorithm with low variance. The problem with this approach is that both steps can be prohibitively expensive. In this paper, we simplify steps (i) and (ii) by calculating only parts of the adjoint solution. More specifically, in a geometry with limited volume scattering and complicated reflection at the boundary, we consider the situation where the adjoint solution "neglects" volume scattering, whereby significantly reducing the degrees of freedom in steps (i) and (ii). A main application for such a geometry is in remote sensing of the environment using physics-based signal models. Volume scattering is then incorporated using an analog sampling algorithm (or more precisely a simple modification of analog sampling called a heuristic sampling algorithm) in order to obtain unbiased estimators. In geometries with weak volume scattering (with a domain of interest of size comparable to the transport mean free path), we demonstrate numerically significant variance reductions and speed-ups (figures of merit).

preprint2010arXiv

Large Deviation Theory for a Homogenized and "Corrected" Elliptic ODE

We study a one-dimensional elliptic problem with highly oscillatory random diffusion coefficient. We derive a homogenized solution and a so-called Gaussian corrector. We also prove a "pointwise" large deviation principle (LDP) for the full solution and approximate this LDP with a more tractable form. These results allow one to access the limits of Gaussian correctors. In general, the corrector does not capture the large deviation behavior. Applications to uncertainty quantification are considered.