Baidu Encyclopedia version
In probability and statistics theory, the generation model refers to a model that can randomly generate observation data, especially given the conditions of certain implicit parameters. It assigns a joint probability distribution to the observations and the annotated data sequence. In machine learning, the generation model can be used to model data directly (for example, to sample data based on a probability density function of a variable), or to establish a conditional probability distribution between variables. The conditional probability distribution can be formed by the generated model according to Bayes' theorem.
In statistical classification, including machine learning, the two main methods are calledgenerateMethods andDiscriminatemethod. These computational classifiers use different methods and the degree of statistical modeling is different. The terms are inconsistent, but can be distinguished by three main types, following Jebara (2004):
- Given an observable variable XAnd target variables ÿ,OneGenerating modelIs a joint probability distribution of a statistical modelX × ÿ.
- ADiscriminant modelYes, the target of the conditional probability of the modelÿGiven the observationXSymbolically
- Classifiers that do not use probability model calculations are also broadly referred to as "discriminability."
The difference between the last two classes is not consistent; Jebara (2004) calls these three categoriesGenerate learning.Conditional learning和Discriminatory learning, but Ng&Jordan (2002) only distinguishes between two categories, calledGenerating classifier(joint distribution) andDiscriminant classifier(conditional distribution or no allocation), there is no distinction between the latter two categories. Similarly, based on the generated modelClassifierBeGenerating classifierAnd the classifier based on the discriminant model isDiscriminant classifier, although the term also refers to a model-based classifier. Each standard example is a linear classifier, which is: Generate Classifier: Naive Bayes Classifier and Linear Discriminant Analysis; Discriminant Model: Logistic Regression; Non-Model Classifier: Perceptron and Support Vector Machine.