Chapter 17: Factor Graphs

Learning Objectives

• Convert a joint distribution into a factor graph and read off the conditional-independence structure
• Identify the factor graph behind HMMs, LDPC codes, MIMO detection, and ISI channels
• Distinguish tractable inference on trees from approximate inference on loopy graphs
• Explain why cycles break exactness and how short cycles degrade loopy BP performance
• Recognize when a problem admits an efficient message-passing solution