Sampling: Two Basic Algorithms
In probabilistic machine learning, exact inference is often difficult or impossible when we cannot sample from or estimate the properties of a particular distribution. While we can leverage conjugacy to keep the distributions tractable or use variational inference to deterministically approximate a density, we can also use randomized methods for inference using numerical sampling, sometimes called Monte Carlo methods.
Introduction In probabilistic machine learning, exact inference is often difficult or impossible when we cannot sample from or estimate the properties of a particular distribution. While we can leverage conjugacy to keep the distributions tractable or use variational inference to deterministically approximate a density, we can also use randomized methods for inference using numerical sampling, sometimes called Monte Carlo methods. While we are sometimes interested in a posterior distribution per se, we are often only interested in properties of the distribution. For example, we need the…
saved by
related reading
- A Gallery of Methods Beyond RL — Part I: Sampling Methodsshengyu-feng.github.io
- [1809.09087] Implicit Maximum Likelihood Estimationarxiv.org
- Coordinate Ascent Mean-field Variational Inference (Univariate Gaussian Example)suzyahyah.github.io
- Approximating KL Divergencejoschu.net
- Bayesian Neural Networkscs.toronto.edu
- Importance Sampling Explained | Built Inbuiltin.com
- Sending Samples Without Bits-Backjoschu.net
- A Simplified Overview of Langevin Dynamicsfriedmanroy.github.io
- Gibbs sampling - Wikipediaen.wikipedia.org
- Competing with sampling — Alignment Research Centeralignment.org
- practicalqmc.pdfartowen.su.domains
- High-Dimensional Gaussian Sampling: A Review and a Unifying Approach Based on a Stochastic Proximal Point Algorithmarxiv.org