14  Machine learning foundations for vision

Chapter status: Structural placeholder.

This chapter currently contains the approved heading structure but no reviewed technical content.

14.1 The learning paradigm

14.1.1 Supervised, unsupervised and self-supervised learning

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.1.2 Training, validation and test sets

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.1.3 Bias, variance and generalisation

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.1.4 Cross-validation and model selection

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.2 Classical classifiers

14.2.1 Linear and logistic regression

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.2.2 K-nearest neighbours

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.2.3 Support vector machines and kernels

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.2.4 Decision trees, random forests and boosting

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.2.5 Bayesian classifiers

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.3 Feature engineering and dimensionality reduction

14.3.1 Hand-crafted image features

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.3.2 Principal component analysis

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.3.3 Linear discriminant analysis

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.3.4 Manifold learning (outline)

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.4 Loss functions and optimisation

14.4.1 Squared, absolute and Huber loss

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.4.2 Cross-entropy and log-loss

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.4.3 Gradient descent and its variants

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.4.4 Regularisation: L1, L2, elastic net

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.5 Evaluation of classifiers

14.5.1 The confusion matrix

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.5.2 Accuracy, precision, recall, F-score

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.5.3 ROC and precision-recall curves; AUC

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.5.4 Class imbalance and industrial reality

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.6 Choosing classical versus deep approaches

14.6.1 Data availability and labelling cost

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.6.2 Interpretability and validation burden

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.

14.6.3 When hand-crafted features still win

Content status: Placeholder.

This subsection has not yet been drafted or technically reviewed.