ML index

Machine learning studies algorithms that improve at a task by fitting patterns in data. Mathematically, most ML is statistics + linear algebra + optimisation + honest evaluation. The source PDF in this folder, ML_cheatsheat, is a classification-vs-regression cheat sheet; these notes rewrite it into searchable Obsidian form and add the missing equations/intuition.

Core route

1. ML Key concepts - what supervised learning is, what classification/regression mean, and how the common model families differ.
2. ML Equations and definitions - formulas for losses, margins, Bayes rule, regularisation, and regression/classification models.
3. ML Models from cheatsheet - every model named in the PDF, with classification/regression usage where relevant.
4. ML Examples - small examples and interpretive checks.
5. ML Common pitfalls - traps to avoid.
6. ML Questions to answer - revision prompts.

Main supervised split

Task	Output	Examples from the PDF	Typical question
Classification	Categorical label or class probability	Logistic regression, LDA, Naive Bayes, SVM, KNN, decision trees, random forests, gradient boosting	”Which class is this?”
Regression	Continuous value, often with residual/error	Linear regression, polynomial regression, Gaussian process regression, SVR, KNN regression, regression trees, forests, boosting, Lasso, Ridge	”What number should this be?”

Models covered by the PDF

• Shared classification/regression families: support vector machines, k-nearest neighbours, decision trees, random forests, gradient boosting, Lasso, Ridge.
• Classification-only section: logistic regression, linear discriminant analysis, Naive Bayes classifier.
• Regression-only section: linear regression, polynomial regression, Gaussian process regression.

Mental model

An ML model is a parameterised function $f_{h}eta(x)$. Training chooses parameters $heta$ that make predictions useful on observed data. The actual goal is low error on new cases from the same process, not perfect training-set memory. That is why train/validation/test splits matter.

Bridges

Stats Key concepts explains likelihood, priors, uncertainty, and residuals. Vector Calculus Key concepts explains gradients used in optimisation. Residuals in NF2 and PINNs is a physics example where neural networks are trained with both data terms and equation residuals.