Expectation–maximization algorithm
Algoritmo esperanza-maximización
- Date
- 1977 · confidence high
- Place
- Cambridge, Massachusetts, United States
- Credited to
- Arthur Dempster, Nan Laird and Donald Rubin
- Field
- Knowledge & symbols
What it is
An iterative method for estimating maximum-likelihood parameters in statistical models with unobserved latent variables, alternating an expectation (E) step and a maximization (M) step. It is used, for example, to fit mixtures of Gaussians.
Published in a 1977 Journal of the Royal Statistical Society paper that unified and named similar methods previously used separately.
What it led to
- A standard tool of unsupervised machine learning
- Practical fitting of mixture models and hidden-variable models in statistics