Annealed Stein Variational Gradient Descent
arXiv:2101.09815
Abstract
Particle based optimization algorithms have recently been developed as sampling methods that iteratively update a set of particles to approximate a target distribution. In particular Stein variational gradient descent has gained attention in the approximate inference literature for its flexibility and accuracy. We empirically explore the ability of this method to sample from multi-modal distributions and focus on two important issues: (i) the inability of the particles to escape from local modes and (ii) the inefficacy in reproducing the density of the different regions. We propose an annealing schedule to solve these issues and show, through various experiments, how this simple solution leads to significant improvements in mode coverage, without invalidating any theoretical properties of the original algorithm.
References in corpus (9)
- Simulated Tempering: A New Monte Carlo Scheme
- SGDR: Stochastic Gradient Descent with Warm Restarts
- Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishing
- Cyclical Stochastic Gradient MCMC for Bayesian Deep Learning
- Message Passing Stein Variational Gradient Descent
- On the geometry of Stein variational gradient descent
- Variational Tempering
- Improving Explorability in Variational Inference with Annealed Variational Objectives
- Kernel Stein Generative Modeling
Cited by in corpus (6)
- Repulsive Deep Ensembles are Bayesian
- Annealed Stein Variational Gradient Descent for Improved Uncertainty Estimation in Full-Waveform Inversion
- Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent
- On Stein Variational Neural Network Ensembles
- Stein Variational Goal Generation for adaptive Exploration in Multi-Goal Reinforcement Learning
- Neural Variational Gradient Descent