Corpus ID: 47015288

A Stein variational Newton method

@inproceedings{Detommaso2018ASV,
  title={A Stein variational Newton method},
  author={Gianluca Detommaso and T. Cui and Y. Marzouk and Robert Scheichl and Alessio Spantini},
  booktitle={NeurIPS},
  year={2018}
}
Stein variational gradient descent (SVGD) was recently proposed as a general purpose nonparametric variational inference algorithm [Liu & Wang, NIPS 2016]: it minimizes the Kullback-Leibler divergence between the target distribution and its approximation by implementing a form of functional gradient descent on a reproducing kernel Hilbert space. [...] Key Method We also show how second-order information can lead to more effective choices of kernel. We observe significant computational gains over the original…Expand
Stein Variational Gradient Descent With Matrix-Valued Kernels
TLDR
This work enhances SVGD by leveraging preconditioning matrices, such as the Hessian and Fisher information matrix, to incorporate geometric information into SVGD updates, and presents a generalization of SVGD that replaces the scalar-valued kernels in vanilla SVGD with more general matrix- valued kernels. Expand
Neural Variational Gradient Descent
TLDR
This work proposes Neural Variational Gradient Descent (NVGD), which is based on parametrizing the witness function of the Stein discrepancy by a deep neural network whose parameters are learned in parallel to the inference, mitigating the necessity to make any kernel choices whatsoever. Expand
SVGD as a kernelized Wasserstein gradient flow of the chi-squared divergence
TLDR
A new perspective on SVGD is introduced that views SVGD as the (kernelized) gradient flow of the chi-squared divergence which exhibits a strong form of uniform exponential ergodicity under conditions as weak as a Poincare inequality and proposes an alternative to SVGD, called Laplacian Adjusted Wasserstein Gradient Descent (LAWGD), that can be implemented from the spectral decomposition of the Laplacan operator associated with the target density. Expand
A stochastic version of Stein Variational Gradient Descent for efficient sampling
TLDR
RBM-SVGD, a stochastic version of Stein Variational Gradient Descent method for efficiently sampling from a given probability measure and thus useful for Bayesian inference is proposed. Expand
On the geometry of Stein variational gradient descent
TLDR
This paper focuses 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, and considers certain nondifferentiable kernels with adjusted tails. Expand
Kernel Selection for Stein Variational Gradient Descent
TLDR
This work extends Kernelized Stein Discrepancy to its multiple kernel view and uses it to construct a general algorithm based on SVGD, which be called Multiple Kernel SVGD (MK-SVGD). Expand
Stein variational gradient descent with local approximations
TLDR
This paper approximate the forward model using a deep neural network which is trained on a carefully chosen training set which also determines the quality of the surrogate, and proposes a general adaptation procedure to refine the local approximation online without destroying the convergence of the resulting SVGD. Expand
Projected Stein Variational Newton: A Fast and Scalable Bayesian Inference Method in High Dimensions
TLDR
A fast and scalable variational method for Bayesian inference in high-dimensional parameter space, which is called projected Stein variational Newton (pSVN) method, and demonstrates fast convergence of the proposed method and its scalability with respect to the number of parameters, samples, and processor cores. Expand
Variance Reduction and Quasi-Newton for Particle-Based Variational Inference
TLDR
It is found that existing ParVI approaches using stochastic gradients converge insufficiently fast under sample quality metrics, and a novel variance reduction and quasiNewton preconditioning framework for ParVIs is proposed, by leveraging the Riemannian structure of the Wasserstein space and advanced RiemANNian optimization algorithms. Expand
Stein variational reduced basis Bayesian inversion
TLDR
A Stein variational reduced basis method (SVRB) to solve large-scale PDE-constrained Bayesian inverse problems and develops an adaptive and goal-oriented model reduction technique based on reduced basis approximations for the evaluation of the potential and its gradient. Expand
...
1
2
3
4
5
...

References

SHOWING 1-10 OF 33 REFERENCES
Stein Variational Gradient Descent as Gradient Flow
  • Q. Liu
  • Computer Science, Mathematics
  • NIPS
  • 2017
TLDR
This paper develops the first theoretical analysis on SVGD, discussing its weak convergence properties and showing that its asymptotic behavior is captured by a gradient flow of the KL divergence functional under a new metric structure induced by Stein operator. Expand
Stein Variational Gradient Descent: A General Purpose Bayesian Inference Algorithm
We propose a general purpose variational inference algorithm that forms a natural counterpart of gradient descent for optimization. Our method iteratively transports a set of particles to match theExpand
Analyzing and Improving Stein Variational Gradient Descent for High-dimensional Marginal Inference
Stein variational gradient descent (SVGD) is a nonparametric inference method, which iteratively transports a set of randomly initialized particles to approximate a differentiable targetExpand
Stein Variational Adaptive Importance Sampling
TLDR
A novel adaptive importance sampling algorithm which incorporates Stein variational gradient decent algorithm (SVGD) with importance sampling (IS), which turns SVGD into a standard IS algorithm, allowing for standard diagnostic and analytic tools of IS to evaluate and interpret the results. Expand
Message Passing Stein Variational Gradient Descent
TLDR
Experimental results show MP-SVGD's advantages of preventing vanishing repulsive force in high-dimensional space over SVGD, and its particle efficiency and approximation flexibility over other inference methods on graphical models. Expand
VAE Learning via Stein Variational Gradient Descent
TLDR
A new method for learning variational autoencoders (VAEs) is developed, based on Stein variational gradient descent, demonstrating the scalability of the model to large datasets. Expand
Riemannian Stein Variational Gradient Descent for Bayesian Inference
TLDR
RSVGD is a Bayesian inference method that generalizes Stein Variational Gradient Descent to Riemann manifold and shows the advantages over SVGD of exploring distribution geometry and the advantages of particle-efficiency, iteration-effectiveness and approximation flexibility over other inference methods on Riem Mann manifolds. Expand
Structured Stein Variational Inference for Continuous Graphical Models
TLDR
A novel distributed inference algorithm for continuous graphical models by extending Stein variational gradient descent to leverage the Markov dependency structure of the distribution of interest and outperforms a variety of baselines including standard MCMC and particle message passing methods. Expand
Stein Variational Autoencoder
A new method for learning variational autoencoders is developed, based on an application of Stein’s operator. The framework represents the encoder as a deep nonlinear function through which samplesExpand
Vprop: Variational Inference using RMSprop
TLDR
This paper presents Vprop, a method for Gaussian variational inference that can be implemented with two minor changes to the off-the-shelf RMSprop optimizer, as a principled, computationally-efficient, and easy-to-implement method for Bayesian deep learning. Expand
...
1
2
3
4
...