Stochastic Generative Process

A Stochastic Generative Process is a generative process that is a stochastic system.

• Context:
  • It can be modeled by a Generative Statistical Model.
• See: Deterministic Generative Process, Predictive Model.

---

---

References

2014

• http://factorie.cs.umass.edu/usersguide/UsersGuide030Overview.html
  • Undirected graphical models (also known as Markov random fields) represent a joint distribution over random variables by a product of unnormalized non-negative values (one value for each clique in the graph). They are convenient models for data in which it is not intuitive to impose an ordering on the variables' generative process.

1999

• (McCallum, 1999) ⇒ Andrew McCallum. (1999). “Multi-label Text Classification with a Mixture Model Trained by EM.” In: AAAI 99 Workshop on Text Learning.
  • QUOTE: We define a probabilistic generative model that represents the multi-label nature of a document by indicating that the words in a document are produced by a mixture of word distributions, one for each topic. The generative process begins by selecting the set of classes (instead of a single class) that will be the labels for this document; then producing a set of mixture weights for those classes; finally, each word in the document is generated by first selecting a class according to these mixture weights, then letting that class generate a single word. Classification uses Bayes rule and selects the class set that is most likely given the document.

---