Piecewise Deterministic Sampling for Constrained Distributions
By: Joël Tatang Demano, Paul Dobson, Konstantinos Zygalakis
Potential Business Impact:
Helps computers learn from data with rules.
In this paper, we propose a novel class of Piecewise Deterministic Markov Processes (PDMP) that are designed to sample from constrained probability distributions $\pi$ supported on a convex set $\mathcal{M}$. This class of PDMPs adapts the concept of a mirror map from convex optimisation to address sampling problems. Such samplers provides unbiased algorithms that respect the constraints and, moreover, allow for exact subsampling. We demonstrate the advantages of these algorithms on a range of constrained sampling problems where the proposed algorithm outperforms state of the art stochastic differential equation-based methods.
Similar Papers
Covariance-Adaptive Bouncy Particle Samplers via Split Lagrangian Dynamics
Computation
Helps computers learn faster by changing how they move.
Towards practical PDMP sampling: Metropolis adjustments, locally adaptive step-sizes, and NUTS-based time lengths
Computation
Makes computer guessing of tricky patterns faster.
Transient regime of piecewise deterministic Monte Carlo algorithms
Computation
Helps computers find answers faster in complex problems.