Model-Based Machine Learning

✍ Scribed by John Michael Winn, Christopher M. Bishop, Thomas Diethe, John Guiver, Yordan Zaykov

• Publisher: CRC Press
• Year: 2024
• Tongue: English
• Leaves: 469
• Category: Library

✦ Synopsis

A fundamental challenge when using machine learning is connecting the abstract mathematics of a machine learning technique to real world problems. This book tackles this challenge through model-based machine learning, focusing on understanding the assumptions encoded in a machine learning system.

✦ Table of Contents

Cover
Half Title
Title Page
Copyright Page
Dedication
Table of Contents
Preface
How can machine learning solve my problem?
Chapter 1 A murder mystery
1.1 INCORPORATING EVIDENCE
1.2 UPDATING OUR BELIEFS
1.2.1 Two rules for working with joint probabilities
1.2.2 Inference using the joint distribution
1.3 A MODEL OF A MURDER
1.3.1 Inference without computing the joint distribution
1.4 EXTENDING THE MODEL
1.4.1 Incremental inference
Chapter 2 Assessing people’s skills
2.1 A MODEL IS A SET OF ASSUMPTIONS
2.1.1 Questioning our assumptions
2.2 TESTING OUT THE MODEL
2.2.1 Doing inference by hand
2.2.2 Doing inference by passing messages on the graph
2.2.3 Using belief propagation to test out the model
2.3 LOOPINESS
2.3.1 Loopy belief propagation
2.3.2 Applying loopy belief propagation to our model
2.4 MOVING TO REAL DATA
2.4.1 Visualising the data
2.4.2 A factor graph for the whole test
2.4.3 Our first results
2.5 DIAGNOSING THE PROBLEM
2.5.1 Checking the inference algorithm
2.5.2 Working out what is wrong with the model
2.6 LEARNING THE GUESS PROBABILITIES
2.6.1 Representing uncertainty in continuous values
2.6.2 Measuring progress
2.6.3 A different way of measuring progress
2.6.4 Finishing up
Interlude: the machine learning life cycle
Chapter 3 Meeting your match
3.1 MODELLING THE OUTCOME OF GAMES
3.1.1 Modelling how well someone plays
3.1.2 Computing the probability of winning
3.2 INFERRING THE PLAYERS’ SKILLS
3.2.1 Modelling skills
3.2.2 Inference in the TrueSkill model
3.2.3 A problem with using exact inference
3.3 A SOLUTION: EXPECTATION PROPAGATION
3.3.1 Applying expectation propagation
3.3.2 Multiple games
3.4 EXTENSIONS TO THE CORE MODEL
3.4.1 What if a game can end in a draw?
3.4.2 What if we have more than two players in a game?
3.4.3 What if the games are played by teams?
3.5 ALLOWING THE SKILLS TO VARY
3.5.1 Reproducing the problem
3.5.2 The final model
Chapter 4  Uncluttering your inbox
4.1 COLLECTING AND MANAGING EMAIL DATA
4.1.1 Learning from confidential data
4.2 A MODEL FOR CLASSIFICATION
4.2.1 A one-feature classification model
4.3 MODELLING MULTIPLE FEATURES
4.3.1 Features are part of the model
4.4 DESIGNING A FEATURE SET
4.4.1 Features with many states
4.4.2 Numeric features
4.4.3 Features with many, many states
4.4.4 An initial feature set
4.5 EVALUATING AND IMPROVING THE FEATURE SET
4.5.1 Parallel and sequential schedules
4.5.2 Visualising the learned weights
4.5.3 Evaluating reply prediction
4.5.4 Understanding the user’s experience
4.5.5 Improving the feature set
4.6 LEARNING AS EMAILS ARRIVE
4.6.1 Modelling a community of users
4.6.2 Solving the cold start problem
4.6.3 Final testing and changes
Chapter 5 Making recommendations
5.1 LEARNING ABOUT PEOPLE AND MOVIES
5.1.1 Characterising movies
5.1.2 A model of a trait
5.2 MULTIPLE TRAITS AND MULTIPLE PEOPLE
5.2.1 Learning from many people at once
5.3 TRAINING OUR RECOMMENDER
5.3.1 Getting to know our data
5.3.2 Training on MovieLens data
5.4 OUR FIRST RECOMMENDATIONS
5.4.1 Evaluating our predictions
5.4.2 How many traits should we use?
5.5 MODELLING STAR RATINGS
5.5.1 Results with star ratings
5.6 ANOTHER COLD START PROBLEM
5.6.1 Adding features to our model
5.6.2 Results with features
5.6.3 Final thoughts
Chapter 6 Understanding asthma
6.1 A MODEL OF ALLERGIES
6.1.1 Modelling test results
6.1.2 Modelling tests through time
6.1.3 Completing the model
6.1.4 Reviewing our assumptions
6.2 TRYING OUT THE MODEL
6.2.1 Working with missing data
6.2.2 Some initial results
6.3 COMPARING ALTERNATIVE MODELS
6.3.1 Comparing the two models using Bayesian model selection
6.4 MODELLING WITH GATES
6.4.1 Using gates for model selection
6.4.2 Expectation propagation in factor graphs with gates
6.5 DISCOVERING SENSITISATION CLASSES
6.5.1 Testing the model with two classes
6.5.2 Exploring more sensitisation classes
Chapter 7 Harnessing the crowd
7.1 A MODEL OF A CROWD WORKER
7.1.1 A simpler setting
7.1.2 Using more than two labels
7.1.3 Incorporating crowd worker labels
7.1.4 Completing the model
7.2 TRYING OUT THE WORKER MODEL
7.3 CORRECTING FOR WORKER BIASES
7.3.1 Evaluating our biased worker model
7.3.2 Comparing more and less flexible models
7.4 COMMUNITIES OF WORKERS
7.4.1 Results of the community model
7.4.2 Results with less training data
7.5 MAKING USE OF THE TWEETS
7.5.1 Results with words
7.5.2 Wrapping up
Chapter 8 How to read a model
8.1 LATENT DIRICHLET ALLOCATION
8.1.1 Exploring the assumptions in LDA
8.1.2 Extensions to LDA
8.2 DECISION TREE
8.2.1 Factor graph for a decision tree
8.2.2 What assumptions are being made?
8.2.3 Decision forest
8.3 PRINCIPAL COMPONENT ANALYSIS
8.3.1 Computing the principal components
8.3.2 A factor graph for PCA
8.3.3 The assumptions built in to PCA
8.3.4 Extensions to PCA
8.4 K-MEANS CLUSTERING
8.4.1 The k-means algorithm
8.4.2 A model for k-means
8.4.3 Some hidden assumptions in k-means
8.4.4 Problems with k-means
Afterword
Bibliography
Index