Ensemble-Based Gradient Inference for Particle Methods in Optimization and Sampling

DOI10.1137/22M1533281zbMath1518.65142arXiv2209.15420MaRDI QIDQ6177924

Philipp Wacker, Claudia Schillings, Claudia Totzeck

Publication date: 31 August 2023

Published in: SIAM/ASA Journal on Uncertainty Quantification (Search for Journal in Brave)

Abstract: We propose an approach based on function evaluations and Bayesian inference to extract higher-order differential information of objective functions {from a given ensemble of particles}. Pointwise evaluation ${V(x^i)}_i$ of some potential $V$ in an ensemble ${x^i}_i$ contains implicit information about first or higher order derivatives, which can be made explicit with little computational effort (ensemble-based gradient inference -- EGI). We suggest to use this information for the improvement of established ensemble-based numerical methods for optimization and sampling such as Consensus-based optimization and Langevin-based samplers. Numerical studies indicate that the augmented algorithms are often superior to their gradient-free variants, in particular the augmented methods help the ensembles to escape their initial domain, to explore multimodal, non-Gaussian settings and to speed up the collapse at the end of optimization dynamics.} The code for the numerical examples in this manuscript can be found in the paper's Github repository (https://github.com/MercuryBench/ensemble-based-gradient.git).

Full work available at URL: https://arxiv.org/abs/2209.15420

zbMATH Keywords

samplingoptimizationLangevin dynamicsensemble methods

Mathematics Subject Classification ID

Probabilistic methods, particle methods, etc. for boundary value problems involving PDEs (65N75) Bayesian inference (62F15) Nonconvex programming, global optimization (90C26) Derivative-free methods and methods using generalized derivatives (90C56) Stochastic ordinary differential equations (aspects of stochastic analysis) (60H10) Dynamical systems in optimization and economics (37N40) Vlasov equations (35Q83)

Cites Work

• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Gradient and diagonal Hessian approximations using quadratic interpolation models and aligned regular bases
• Exponential convergence of Langevin distributions and their discrete approximations
• A note on the mean-field limit for the particle swarm optimization
• Inexact derivative-free optimization for bilevel learning
• Consensus-based global optimization with personal best
• Iterative ensemble Kalman methods: a unified perspective with some new variants
• Optimal scaling of random-walk Metropolis algorithms on general target distributions
• A derivative-free Gauss-Newton method
• The calculus of simplex gradients
• Efficient calculation of regular simplex gradients
• Geometry of interpolation sets in derivative free optimization
• On the Lagrange functions of quadratic models that are defined by interpolation^*
• A consensus-based model for global optimization and its mean-field limit
• Locally Weighted Regression: An Approach to Regression Analysis by Local Fitting
• Consensus-based optimization on hypersurfaces: Well-posedness and mean-field limit
• Using simplex gradients of nonsmooth functions in direct search methods
• Introduction to Derivative-Free Optimization
• Robust Locally Weighted Regression and Smoothing Scatterplots
• The Variational Formulation of the Fokker--Planck Equation
• An analytical framework for consensus-based global optimization method
• Affine Invariant Interacting Langevin Dynamics for Bayesian Inference
• A consensus-based global optimization method for high dimensional machine learning problems
• Trends in Consensus-Based Optimization
• Stochastic consensus dynamics for nonconvex optimization on the Stiefel manifold: Mean-field limit and convergence
• Interacting Langevin Diffusions: Gradient Structure and Ensemble Kalman Sampler
• From particle swarm optimization to consensus based optimization: Stochastic modeling and mean-field limit
• Well posedness and convergence analysis of the ensemble Kalman inversion
• Analysis of the Ensemble Kalman Filter for Inverse Problems
• Consensus-based optimization via jump-diffusion stochastic differential equations
• An adaptive consensus based method for multi-objective optimization with uniform Pareto front approximation

Related Items (2)

EnKSGD: a class of preconditioned black box optimization and inversion algorithms ⋮ Leveraging memory effects and gradient information in consensus-based optimisation: on global convergence in mean-field law

This page was built for publication: Ensemble-Based Gradient Inference for Particle Methods in Optimization and Sampling