Optimal Preconditioning and Fisher Adaptive Langevin Sampling

by   Michalis K. Titsias, et al.

We define an optimal preconditioning for the Langevin diffusion by analytically optimizing the expected squared jumped distance. This yields as the optimal preconditioning an inverse Fisher information covariance matrix, where the covariance matrix is computed as the outer product of log target gradients averaged under the target. We apply this result to the Metropolis adjusted Langevin algorithm (MALA) and derive a computationally efficient adaptive MCMC scheme that learns the preconditioning from the history of gradients produced as the algorithm runs. We show in several experiments that the proposed algorithm is very robust in high dimensions and significantly outperforms other methods, including a closely related adaptive MALA scheme that learns the preconditioning with standard adaptive MCMC as well as the position-dependent Riemannian manifold MALA sampler.


Kernel Adaptive Metropolis-Hastings

A Kernel Adaptive Metropolis-Hastings algorithm is introduced, for the p...

Adaptive Designs for Optimal Observed Fisher Information

Expected Fisher information can be found a priori and as a result its in...

Adaptive Schemes for Piecewise Deterministic Monte Carlo Algorithms

The Bouncy Particle sampler (BPS) and the Zig-Zag sampler (ZZS) are cont...

Convergence of position-dependent MALA with application to conditional simulation in GLMMs

We establish verifiable conditions under which Metropolis-Hastings (MH) ...

Deep Latent Dirichlet Allocation with Topic-Layer-Adaptive Stochastic Gradient Riemannian MCMC

It is challenging to develop stochastic gradient based scalable inferenc...

Theoretical foundation for CMA-ES from information geometric perspective

This paper explores the theoretical basis of the covariance matrix adapt...

Please sign up or login with your details

Forgot password? Click here to reset