On the geometry of Stein variational gradient descent
arXiv:1912.00894
Abstract
Bayesian inference problems require sampling or approximating high-dimensional probability distributions. The focus of this paper is on the recently introduced Stein variational gradient descent methodology, a class of algorithms that rely on iterated steepest descent steps with respect to a reproducing kernel Hilbert space norm. This construction leads to interacting particle systems, the mean-field limit of which is a gradient flow on the space of probability distributions equipped with a certain geometrical structure. We leverage this viewpoint to shed some light on the convergence properties of the algorithm, in particular addressing the problem of choosing a suitable positive definite kernel function. Our analysis leads us to considering certain nondifferentiable kernels with adjusted tails. We demonstrate significant performance gains of these in various numerical experiments.
arXiv comment40 pages, 4 figures
References in corpus (19)
- The Ensemble Kalman Filter for Inverse Problems
- Stein Variational Gradient Descent as Gradient Flow
- Maximum Mean Discrepancy Gradient Flow
- Message Passing Stein Variational Gradient Descent
- Stein Variational Gradient Descent With Matrix-Valued Kernels
- Understanding and Accelerating Particle-Based Variational Inference
- A stochastic version of Stein Variational Gradient Descent for efficient sampling
- Projected Stein Variational Newton: A Fast and Scalable Bayesian Inference Method in High Dimensions
- Variational Modelling: Energies, gradient flows, and large deviations
- Stein Variational Gradient Descent as Moment Matching
- Stochastic Gradient MCMC with Repulsive Forces
- Accelerating Langevin Sampling with Birth-death
- Dynamics of a planar Coulomb gas
- Greedy inference with structure-exploiting lazy maps
- Note on Interacting Langevin Diffusions: Gradient Structure and Ensemble Kalman Sampler by Garbuno-Inigo, Hoffmann, Li and Stuart
- Kernel embedding of maps for sequential Bayesian inference: The variational mapping particle filter
- Particle Optimization in Stochastic Gradient MCMC
- Stein Variational Online Changepoint Detection with Applications to Hawkes Processes and Neural Networks
- Wasserstein variational gradient descent: From semi-discrete optimal transport to ensemble variational inference
Cited by in corpus (18)
- A Non-Asymptotic Analysis for Stein Variational Gradient Descent
- Repulsive Deep Ensembles are Bayesian
- Birth-death dynamics for sampling: Global convergence, approximations and their asymptotics
- Annealed Stein Variational Gradient Descent
- Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent
- On Stein Variational Neural Network Ensembles
- Kernel Stein Discrepancy Descent
- Convergence of Stein Variational Gradient Descent under a Weaker Smoothness Condition
- Stein Variational Gaussian Processes
- Rough McKean-Vlasov dynamics for robust ensemble Kalman filtering
- The Wasserstein Proximal Gradient Algorithm
- On the Convergence of Gradient Descent in GANs: MMD GAN As a Gradient Flow
- Variational Transport: A Convergent Particle-BasedAlgorithm for Distributional Optimization
- The equivalence between Stein variational gradient descent and black-box variational inference
- On anisotropic diffusion equations for label propagation
- Relative Entropy Gradient Sampler for Unnormalized Distributions
- Multilevel Stein variational gradient descent with applications to Bayesian inverse problems
- Particle Dynamics for Learning EBMs