liveBook · Manning

Message passing

As explained in Timeless Algorithms, The Foundational Ideas

Message passing is a procedure that propagates information between connected nodes to compute or approximate quantities in a network.

Message passing makes global inference manageable by distributing calculations across local relationships. In a probabilistic graphical model, variable nodes represent observed or hidden quantities, and factor nodes represent probabilistic relationships. Each node combines incoming probability summaries with its own information, then sends updated messages to neighboring nodes. section 10.3.2

From decoding to inference

In the correspondence between decoding and graphical models, unknown bits become hidden variables, coding constraints become probabilistic factors, and decoding messages become local belief updates. On graphs without loops, these updates can produce exact marginal probabilities; with loops, they are generally approximate and may not converge. section 10.3.2

How Timeless Algorithms, The Foundational Ideas covers it

How deep it goes
Defines local belief updates and distinguishes exact marginals on loop-free graphs from approximate, potentially nonconvergent updates on loopy graphs.
Examples
Maps unknown bits to hidden variables, coding constraints to probabilistic factors, and decoding messages to local belief updates.
What you can do after
Understand how local probability summaries propagate evidence and when the resulting inference is exact or approximate.
How it connects
Connects belief propagation, probabilistic graphical models, decoding, and Bayesian inference.

Also explained in

Compare how 13 books explain message passing →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Timeless Algorithms, The Foundational Ideas. The section links point to the parts of the book it draws on.