Accelerated information gradient flow
From MaRDI portal
Publication:2053344
DOI10.1007/S10915-021-01709-3zbMATH Open1484.90059arXiv1909.02102OpenAlexW2971906246MaRDI QIDQ2053344FDOQ2053344
Authors: Wuchen Li, Yifei Wang 
Publication date: 29 November 2021
Published in: Journal of Scientific Computing (Search for Journal in Brave)
Abstract: We present a framework for Nesterov's accelerated gradient flows in probability space to design efficient mean-field Markov chain Monte Carlo (MCMC) algorithms for Bayesian inverse problems. Here four examples of information metrics are considered, including Fisher-Rao metric, Wasserstein-2 metric, Kalman-Wasserstein metric and Stein metric. For both Fisher-Rao and Wasserstein-2 metrics, we prove convergence properties of accelerated gradient flows. In implementations, we propose a sampling-efficient discrete-time algorithm for Wasserstein-2, Kalman-Wasserstein and Stein accelerated gradient flows with a restart technique. We also formulate a kernel bandwidth selection method, which learns the gradient of logarithm of density from Brownian-motion samples. Numerical experiments, including Bayesian logistic regression and Bayesian neural network, show the strength of the proposed methods compared with state-of-the-art algorithms.
Full work available at URL: https://arxiv.org/abs/1909.02102
Recommendations
- Is there an analog of Nesterov acceleration for gradient-based MCMC?
- Natural gradient via optimal transport
- The Bayesian update: variational formulations and gradient flows
- A note on parametric Bayesian inference via gradient flows
- Projected Wasserstein Gradient Descent for High-Dimensional Bayesian Inference
Cites Work
- Functional and shape data analysis
- Adaptive restart for accelerated gradient schemes
- Title not available (Why is that?)
- THE GEOMETRY OF DISSIPATIVE EVOLUTION EQUATIONS: THE POROUS MEDIUM EQUATION
- Inverse problems: a Bayesian perspective
- Title not available (Why is that?)
- Title not available (Why is that?)
- The Density Manifold and Configuration Space Quantization
- Improvement on some known nonparametric uniformly consistent estimators of derivatives of a density
- Title not available (Why is that?)
- Wasserstein Riemannian geometry of Gaussian densities
- On Wasserstein geometry of the space of Gaussian measures
- Information geometry and its applications
- Convergence to equilibrium in Wasserstein distance for damped Euler equations with interaction forces
- A variational perspective on accelerated methods in optimization
- Geometry of matrix decompositions seen through optimal transport and information geometry
- A blob method for diffusion
- Interacting Langevin diffusions: gradient structure and ensemble Kalman sampler
- Is there an analog of Nesterov acceleration for gradient-based MCMC?
- Wasserstein Hamiltonian flows
- A Mean-Field Optimal Control Formulation for Global Optimization
Cited In (13)
- Hölder stability and uniqueness for the mean field games system via Carleman estimates
- A Time-Fractional Mean-Field Control Modeling Subdiffusive Advective Transport
- A note on parametric Bayesian inference via gradient flows
- Stein variational gradient descent on infinite-dimensional space and applications to statistical inverse problems
- Global Optimization via Schrödinger–Föllmer Diffusion
- Principal whitened gradient for information geometry
- Is there an analog of Nesterov acceleration for gradient-based MCMC?
- High order spatial discretization for variational time implicit schemes: Wasserstein gradient flows and reaction-diffusion systems
- Noise-free sampling algorithms via regularized Wasserstein proximals
- The Bayesian update: variational formulations and gradient flows
- Sparse optimization on measures with over-parameterized gradient descent
- Natural gradient via optimal transport
- Scaling Up Bayesian Uncertainty Quantification for Inverse Problems Using Deep Neural Networks
Uses Software
This page was built for publication: Accelerated information gradient flow
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2053344)
