No. 263 of 500 No. 49 in Machine Learning

Probabilistic Graphical Models

Principles and Techniques

By Daphne Koller and Nir Friedman

Published
July 31, 2009
Print length
1231 pages
Publisher
The MIT Press
Reading level
Advanced
ISBN-13
978-0262013192
Written for
Graduate students, researchers and ML engineers who need rigorous foundations in probabilistic reasoning and inference.

Synopsis

Koller and Friedman build a unified framework for representing uncertainty with graphs, in which nodes stand for random variables and edges encode the dependencies among them. The book moves through directed Bayesian networks, undirected Markov networks, Gaussian and hybrid models, and template-based extensions for temporal and relational data, and it treats every model class through the same three questions: how to represent a distribution compactly, how to compute answers from it, and how to learn its parameters and structure from data. A final part turns to causality and decision making, including influence diagrams and the value of information.

The approach is mathematical and exhaustive. Exact inference by variable elimination and clique trees sits beside approximate methods such as loopy belief propagation, variational techniques and Markov chain Monte Carlo sampling, and a reader's guide in the introduction maps routes through the material for courses of different lengths. Readers who work through the relevant parts gain a durable vocabulary for probabilistic reasoning that underlies causal inference, probabilistic programming and the latent-variable ideas inside many generative models, though the book itself predates the deep learning era and says little about neural networks.

Scorecard

Editorial score

How it scored

Editorial score: 76.3 / 1005 weighted criteria

  • Influence30%

    27.0/30
  • Authority25%

    23.8/25
  • Practical value20%

    12.0/20
  • Currency15%

    9.0/15
  • Readability10%

    4.5/10
  • Editorial score

    76.3/100

How the rank was computed

Rank No. 263 of 500 comes from a composite of 54.2: observable signals 19.8 of 55 points plus editorial review 34.4 of 45 points.

All 11 inputs
InputPoints
Observable signals19.8/55
Reach (Amazon ratings volume)7.7/16.5
Consensus across best-of lists5.5/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.8/8.3
Amazon sales rank0.8/5.5
Recency0.0/2.8
Editorial review34.4/45
Author authority10.7/11.3
Influence12.2/13.5
Practical value5.4/9.0
Readability2.0/4.5
Currency4.1/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

The most complete textbook on Bayesian and Markov networks, rigorous and authoritative, but its 1,200 pages demand a committed graduate-level reader.

Strengths

  • Unified representation, inference and learning framework
  • Exhaustive coverage of exact and approximate inference
  • Reader's guide supports courses of varied length

Keep in mind

Its length, notation and price make it a poor first introduction, and readers seeking deep learning methods will find little modern neural network material.

Key takeaways

  1. Graphical models make high-dimensional probability tractable by encoding conditional independencies, so a joint distribution factors into small, interpretable local pieces.
  2. Bayesian networks suit directed, cause-and-effect dependencies, while Markov networks capture symmetric interactions; the book shows when each representation fits and how to convert between them.
  3. Exact inference is computationally hard in general, so practitioners choose among clique trees, belief propagation, variational methods and sampling according to model structure and accuracy needs.
  4. Learning divides into parameter estimation and structure search, with expectation maximization handling missing data and Bayesian scores balancing model fit against complexity.

Notable facts

  • Stanford's CS 228 course on probabilistic graphical models listed the book as its required textbook. Source (opens in a new tab)
  • Reviewing it in The Knowledge Engineering Review (2011), Simon Parsons wrote that no earlier book on graphical models could 'hold a candle' to it and named its sheer size as its only problem. Source (opens in a new tab)
  • Koller's Stanford Online specialization on Coursera, split into Representation, Inference and Learning courses, lists about 99,800 learners across its three courses. Source (opens in a new tab)

About the author

Daphne Koller and Nir Friedman

Daphne Koller is a longtime Stanford computer science professor, a MacArthur Fellow and a member of the National Academy of Engineering and the National Academy of Sciences. She co-founded Coursera in 2012 and founded the machine learning drug-discovery company insitro in 2018. Co-author Nir Friedman is a professor of computer science and life sciences at the Hebrew University of Jerusalem, known for applying Bayesian networks to gene expression data.

Questions

Is Probabilistic Graphical Models worth reading in 2026?
Yes, for readers who need rigorous probabilistic reasoning. Deep learning now dominates applied AI, but the book's treatment of conditional independence, inference and structure learning still underpins causal inference, probabilistic programming and Bayesian methods. Readers should treat it as a reference to study in parts and pair it with a modern deep learning text for neural network methods.
Who should read Probabilistic Graphical Models?
Graduate students, researchers and machine learning engineers comfortable with probability, linear algebra and algorithms will benefit most. It suits anyone building systems that must reason under uncertainty, such as medical diagnosis, genomics or sensor fusion. Newcomers without mathematical training should start with a shorter introduction and return to Koller and Friedman for depth on specific topics.
How long is Probabilistic Graphical Models, and how should I read it?
The MIT Press edition runs to roughly 1,230 pages. The authors include a reader's guide that maps routes through the chapters for different courses and goals, so most readers cover representation and inference first and then select the learning or decision-making chapters they need. Working the exercises and following Koller's online courses helps with the denser material.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.