No. 146 of 500 No. 24 in Machine Learning

Machine Learning

A Probabilistic Perspective

By Kevin P. Murphy

Print length
1104 pages
Publisher
The MIT Press
Reading level
Advanced
ISBN-13
978-0262018029
Published
August 24, 2012
Written for
Graduate students, researchers and ML engineers who want a rigorous probabilistic reference across classical methods.

Synopsis

Kevin Murphy's textbook, more than 1,000 pages long, organizes machine learning around a single idea: that learning is best framed as probabilistic modeling and inference. Rather than presenting a catalog of unrelated algorithms, it builds from probability, statistics, linear algebra and optimization to Bayesian and frequentist estimation, linear and logistic regression, generative classifiers, mixture models and the EM algorithm, latent variable models, kernels and Gaussian processes, directed and undirected graphical models, hidden Markov and state-space models, and approximate inference through variational methods and Monte Carlo sampling. Later chapters introduce sparse linear models, clustering, structure learning and the deep learning of the early 2010s.

The approach rewards readers who want to see how methods connect: many models are specified in the compact language of graphical models, and the book's figures can be regenerated with Murphy's open MATLAB toolkit, PMTK. Breadth is the book's distinguishing feature, and it remains a dependable reference for classical probabilistic methods. Its density makes it hard going as a first text, a common complaint from students, and its neural network material predates transformers. Murphy reworked and updated the content in Probabilistic Machine Learning: An Introduction (2022) and Probabilistic Machine Learning: Advanced Topics (2023), which most new readers should consider first.

Scorecard

Editorial score

How it scored

Editorial score: 74.8 / 1005 weighted criteria

  • Influence30%

    25.5/30
  • Authority25%

    22.5/25
  • Practical value20%

    13.0/20
  • Currency15%

    8.3/15
  • Readability10%

    5.5/10
  • Editorial score

    74.8/100

How the rank was computed

Rank No. 146 of 500 comes from a composite of 59.8: observable signals 26.1 of 55 points plus editorial review 33.7 of 45 points.

All 11 inputs
InputPoints
Observable signals26.1/55
Reach (Amazon ratings volume)9.4/16.5
Consensus across best-of lists5.5/13.8
Awards and bestseller evidence5.0/8.3
Reader rating, sample-adjusted5.4/8.3
Amazon sales rank0.8/5.5
Recency0.0/2.8
Editorial review33.7/45
Author authority10.1/11.3
Influence11.5/13.5
Practical value5.9/9.0
Readability2.5/4.5
Currency3.7/6.8

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

A comprehensive, rigorous probabilistic treatment of machine learning, prized as a reference and now partly superseded by Murphy's newer volumes.

Strengths

  • Exceptional breadth across probabilistic ML methods
  • Unifying model-based framework throughout
  • Reproducible figures backed by open code

Keep in mind

Dense and demanding as a first text, and its deep learning material predates transformers; Murphy's 2022 and 2023 successors cover newer methods.

Key takeaways

  1. Most machine learning methods can be understood as a probabilistic model plus an inference procedure, which makes assumptions explicit and comparisons between methods easier.
  2. Bayesian reasoning provides principled ways to quantify uncertainty, regularize models and choose among them, rather than relying on ad hoc heuristics.
  3. Graphical models offer a compact visual language for specifying complex models, from hidden Markov models to conditional random fields.
  4. Exact inference is often intractable, so variational methods and Monte Carlo sampling are essential tools for fitting realistic models.

Notable facts

About the author

Kevin P. Murphy

Kevin P. Murphy is a principal research scientist and director at Google DeepMind, where he leads a team working on generative models, reinforcement learning, robotics and Bayesian inference. He was an associate professor of computer science and statistics at the University of British Columbia, holds a PhD from UC Berkeley, and served as co-editor-in-chief of the Journal of Machine Learning Research from 2014 to 2017.

Questions

Is Machine Learning: A Probabilistic Perspective still worth reading in 2026?
As a reference for classical probabilistic methods, yes. Its treatment of Bayesian inference, graphical models, mixture models and approximate inference remains sound. Its deep learning material is dated, however, and Murphy has since updated the content in Probabilistic Machine Learning: An Introduction (2022) and Advanced Topics (2023). Most new readers should start with those and keep the 2012 book for its breadth.
Who should read Machine Learning: A Probabilistic Perspective?
It suits graduate students, researchers and practitioners comfortable with calculus, linear algebra and probability who want a unified, model-based view of machine learning. The book reviews the needed background, but its pace and density make it hard going for self-taught beginners. Those readers usually do better with a gentler introduction first, returning to Murphy later as a reference.
What do you need to know before reading Machine Learning: A Probabilistic Perspective?
The book assumes comfort with multivariable calculus, linear algebra and basic probability, roughly the level of an advanced undergraduate in a quantitative field. It includes refresher chapters on probability and optimization, but they move quickly. Programming experience helps, since the accompanying PMTK toolkit is written in MATLAB, while the code for Murphy's later books uses Python.

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.