Undirected Graphical Models for Text & Image

Posted in Science on August 22, 2008


Undirected Graphical Models for Text & Image

Lecture slides:

  • Undirected Graphical Models for Test and Images
  • Overview
  • The Vector Space Model
  • The Corpora Matrix
  • Feature Vector Representation
  • Motivation for modeling latent t opical aspects
  • Latent Semantic Indexing
  • Singular Value Decomposition
  • Latent Semantic Structure
  • Two layer Bayesian network
  • Probabilistic LSI
  • Comparison of model semantics
  • Properties of Directed Networks (1)
  • Properties of Directed Networks (2)
  • Two-layer Markov Random Fields
  • The Undirected Alternative
  • Properties of Harmoniums
  • A Constructive Definition (1)
  • A Constructive Definition (2)
  • The Computational Trade-off
  • A Harmonium for IR
  • A Binomial Word-count Model
  • Comparison of model semantics
  • Comparison of text-models
  • Comparison of topic representation
  • Multi-Source Data
  • GM-mixture
  • The Harmonium Counterpart
  • Inter-Source Associations
  • Multi-wing Harmoniums
  • Examples of Latent Topics
  • Classification
  • Retrieval
  • Precision-recall
  • Annotation
  • Conclusions
  • Conclusions, con'd

Author: Eric Xing, Carnegie Mellon University

Watch Video

Tags: Science, Lectures, Computer Science, Machine Learning, VideoLectures.Net, Graphical Models