This section discusses the standard linear factor analysis model and its non-Gaussian, nonlinear and dynamical extensions, as well as the algorithms proposed in the literature for learning these models.
All factor analysis models aim at finding the underlying factors which have generated the observations. Usually the factors are considered to be real valued. The models can be used, for example, for predicting future observations or analysing the nature of the factors or dependences between observations.
The linear model is computationally efficient but can capture only part of the structure in the observations. The extensions aim at capturing more of the structure without overly compromising the computational efficiency.