Generative Latent Space Models for Text and Image
Lecture slides:
- Outline
- Apoptosis + Medicine
- Probabilistic Graphical Model Primer
- What is a graphical model?
- Recap of Basic Prob. Concepts
- Dependencies among variables
- Graphical Models
- Graphical Models, con'd
- Two types of GMs
- Directed Graphical Models
- Bayesian Network
- Conditional probability tables (CPTs)
- Conditional probability density func. (CPDs)
- Markov Random Fields
- An (incomplete) genealogy of graphical models
- GM application: Speech recognition
- GM application: Evolution
- GM application: Solid State Physics
- Probabilistic Inference
- Learning Graphical Models
- A GM course
- The Problem
- Modeling document collections
- Probabilistic Modeling of Text Documents
- Questions of Interest
- Connecting Probability Models to Data
- Conditionally Independent Observations
- “Plate” Notation
- Example: Gaussian Model
- Example: Bayesian Gaussian Model
- Latent Semantic Structure
- GENERATIVE PROCESS
- A generative model for documents
- Probabilistic LSI
- Latent Dirichlet Allocation
- LDA
- Correlated Topic Model
- Inference Tasks
- Bayesian inference
- Approximate Inference
- Bayesian model selection
- The desired LL curve
- Integrating Topics and Syntax
- Topic Hierarchies
- Modeling Topic Evolution
- Latent Space Models for Images
- Image representation
- Exchangeability
- Corr-LDA
- Automatic annotation
- Text-based image retrieval
- Appendix: approximate inference
Author: Eeric Xing, Carnegie Mellon University