Preface
Introduction to the book, who it is for and how to read it.
How can machine learning solve my problem?
What model-based machine learning is and how it helps solve problems.
1. A Murder Mystery
Introduces all the essential concepts of model-based machine learning, in the course of solving a murder.
probability, random variable, probabilistic inference, probabilistic model, factor graph, Bayes' theorem.
2. Assessing People's Skills
A first application of model-based machine learning: assessing what skills a person has based on their answers in a test.
message passing algorithm, loopy belief propagation, visualisation, evaluation metric, ROC curve.
Interlude: the machine learning life cycle
The typical steps in solving any machine learning problem.
3. Meeting Your Match
A real-world application of model-based machine learning to the problem of matching players in online games.
Gaussian distribution, variance, conjugate distribution, expectation propagation, online learning.
4. Uncluttering Your Inbox
A model that removes the clutter from a user's inbox by learning which emails they are likely to ignore.
overfitting, anonymisation, classification, feature set, precision-recall curve, cold start problem.
5. Making Recommendations
Learning a model of people and movies, so they can be matched together to make useful recommendations.
collaborative filtering, symmetries, symmetry breaking.
6. Understanding Asthma
Modelling the way children acquire allergies, to understand and predict childhood asthma.
time series, missing data, model selection, model evidence, Occam's razor, gate, discrete distribution.
7. Harnessing the Crowd
A model that uses crowd-sourced labels to provide accurate information in crisis situations.
Dirichlet distribution, confusion matrix, naive Bayes classifier.
8. How to Read a Model
Exploring models created by other people to understand the assumptions they are making.
Latent Dirichlet Allocation, decision tree, principal component analysis, k-means clustering.
Afterword
Some final thoughts on the future of model-based machine learning.
©2013-23 Winn, MBML v1.0.