Stochastic counterpart of Nesterov's method for sampling and provable non-acceleration of samplers with heavy-ball structure
A. Sharma
Abstract
Polyak's heavy-ball method and Nesterov's method are two momentum-based methods that have inspired many optimization techniques used in applied fields, including machine learning. Momentum also provides a connection with sampling, and interestingly, many kinetic Langevin samplers reduce to the heavy-ball recurrence when velocity is eliminated and noise is removed. This observation motivates a natural question: what is the corresponding sampler for Nesterov's method? We propose a sampler whose deterministic part is exactly Nesterov's recurrence, driven by temporally correlated Gaussian noise and requires one new gradient evaluation per step. Noise correlation provides O(h2) bias for smooth observables. We establish geometric ergodicity for possibly non-convex potentials. For strongly convex potentials with Lipschitz continuous gradients, we obtain contraction rates and show that, up to an additive constant, the energy decays at a rate of order κ-1/2, where κ is the condition number. For Gaussian targets, we show that relaxation has square-root dependence on κ. We further prove that any sampler based on heavy-ball dynamics with additive noise cannot achieve accelerated convergence uniformly over the class of strongly convex functions. Finally, we compare the proposed sampler with OBABO, BAOAB, HMC, MALA, BPS, and NUTS in numerical experiments and demonstrate the better performance of proposed sampler.
Create a lesson
Related papers
A time-domain adversarial example demonstrating the failure of H1 elements in a-ϕ electromagnetic simulation
Timothy Carlson, Sanjay Govindjee
Signed QoI Gaps for Goal-Oriented Error Estimation: Reliability and Sharpness from the Same First-Order System Residuals
Shun Zhang
Perturbation Algorithm and Iterative Realization of a Four-Layer Semi-Discrete Scheme for an Evolutionary Problem with a Variable Operator
David Gulua, Jemal Rogava, Zurab Vashakidze
Regularity transfer for the Ritz projection
Lars Diening, Julian Rolfes, Abner J. Salgado
High-Order Area-Preserving Runge--Kutta Methods for Surface Diffusion Flow of Curves
Lei Cui, Gengen Zhang
Time-independent equivalent matrix representation for time-domain boundary element analysis of large-scale 3D acoustic problems
Weiyu Zhou, Yongqiang Chen