A Tale of Three Probabilistic Families: Discriminative, Descriptive and Generative Models

  • 2018-10-09 21:54:54
  • Ying Nian Wu, Ruiqi Gao, Tian Han, Song-Chun Zhu
  • 16

Abstract

The pattern theory of Grenander is a mathematical framework where thepatterns are represented by probability models on random variables of algebraicstructures. In this paper, we review three families of probability models,namely, the discriminative models, the descriptive models, and the generativemodels. A discriminative model is in the form of a classifier. It specifies theconditional probability of the class label given the input signal. Thedescriptive model specifies the probability distribution of the signal, basedon an energy function defined on the signal. A generative model assumes thatthe signal is generated by some latent variables via a transformation. We shallreview these models within a common framework and explore their connections. Weshall also review the recent developments that take advantage of the highapproximation capacities of deep neural networks.

 

Quick Read (beta)

loading the full paper ...