Discriminative Statistical Model Structure

A Discriminative Statistical Model Structure is a statistical model structure that was produced by a discriminative modeling system (which accepts a discriminative model family).

• AKA: Discriminatively Trained Model.
• Context:
  • It can range from being a Discriminative Classification Model, to being a Discriminative Ranking Model, to being a Discriminative Estimation Model.
    • It can be a Discriminative Classifier or a Discriminative Estimator.
  • It can (typically) Optimize the Posterior Probability ([math]\displaystyle{ p(\bf{Y}|\bf{X}) }[/math]) for some associated Training Data.
    • It can estimate a Class Conditional Probability Function.
  • It cannot, unlike a Generative Model Instance, be used for a Data Generation Task, because it does not model a Joint Distribution ([math]\displaystyle{ P(t) }[/math]).
• Example(s):
  • a Logistic Regression Model Instance.
  • a Latent Variable Discriminative Model.
  • …
• Counter-Example(s):
  • a Generative Predictive Model/Generative Model Instance, such as a Naive Bayes Model Instance.
• See: Conditional Likelihood, Discriminative Learning Algorithm.

---

---

References

2009

• (McCallum et al., 2009) ⇒ Andrew McCallum, Karl Schultz, and Sameer Singh. (2009). “FACTORIE: Probabilistic Programming via Imperatively Defined Factor Graphs..” In: Advances in Neural Information Processing Systems 22 (NIPS 2009).
  • QUOTE: Discriminatively trained undirected graphical models have had wide empirical success, and there has been increasing interest in toolkits that ease their application to complex relational data.

---