Directional Metropolis-Hastings
Abhirup Mallik, Galin L. Jones

TL;DR
The paper introduces Directional Metropolis-Hastings, a new MCMC kernel with state-dependent covariance that uses target distribution derivatives to improve proposal plausibility, with proven ergodicity and adaptive updates.
Contribution
It presents a novel proposal kernel for Metropolis-Hastings that adapts orientation based on target derivatives, along with ergodicity conditions and adaptive variance schemes.
Findings
Enhanced proposal plausibility through directional adaptation.
Proven geometric ergodicity under specified conditions.
Improved performance in Bayesian generalized linear models.
Abstract
We propose a new kernel for Metropolis Hastings called Directional Metropolis Hastings (DMH) with multivariate update where the proposal kernel has state dependent covariance matrix. We use the derivative of the target distribution at the current state to change the orientation of the proposal distribution, therefore producing a more plausible proposal. We study the conditions for geometric ergodicity of our algorithm and provide necessary and sufficient conditions for convergence. We also suggest a scheme for adaptively update the variance parameter and study the conditions of ergodicity of the adaptive algorithm. We demonstrate the performance of our algorithm in a Bayesian generalized linear model problem.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMarkov Chains and Monte Carlo Methods · Bayesian Methods and Mixture Models · Statistical Methods and Inference
