From Artificial Intelligence to Machine Learning
A rigorous, progressive curriculum for a computer-science graduate — built on the principle of understanding why before how. We travel from the definition of AI to implementing core ML algorithms from scratch, and we deliberately stop at the edge of Deep Learning.
How to read this roadmap
Lessons are ordered so that each concept is a prerequisite for the next. Badges tell you what to expect: Ready means the lesson is written and linked; From scratch means we build the algorithm in raw NumPy before touching Scikit-learn; Planned means it's on the path and I'll generate it when you reach it.
What you'll be able to do at the end
- Explain what AI is, how the field developed, and precisely where ML sits inside it.
- Describe the major approaches to building intelligent systems and why statistical learning came to dominate.
- Reason about the ML core: models, parameters, loss functions, and optimization.
- Prepare data, train models, evaluate them honestly, and interpret the results.
- Implement Linear Regression, Logistic Regression, k-NN, Decision Trees, and k-Means from scratch.
- Choose the right algorithm and evaluation strategy for a given problem.
- Walk into your university AI/ML courses already fluent in the vocabulary and intuition.
Part I · The AI Landscape
Before any algorithm, you need the map: what AI is, the approaches people have tried, and where Machine Learning fits.
Part II · The Machine Learning Mindset
The conceptual core that every algorithm later reuses — plus just enough tooling and math to be dangerous.
Learning From Data
Core conceptsTools & Just-Enough Math
SetupPart III · Supervised Learning — Regression
Predicting numbers. We start here because Linear Regression is the cleanest possible instance of the whole ML machine.
Regression
Supervised- 10Linear Regression — the reasoning, the math, the geometryFrom scratch
- 11Gradient Descent in depth — and implement it in NumPyFrom scratch
- 12Multiple & polynomial regression, feature scalingFrom scratch
- 13Overfitting, the bias–variance trade-off, regularization (Ridge/Lasso)Planned
Part IV · Supervised Learning — Classification
Predicting categories. The workhorse family of classical ML — five algorithms, each teaching a different idea.
Classification
Supervised- 14The classification problem & Logistic RegressionFrom scratch
- 15k-Nearest Neighbors — the laziest learnerFrom scratch
- 16Decision Trees — learning if/else from dataFrom scratch
- 17Naive Bayes — classification with probabilityPlanned
- 18Support Vector Machines — margins & the kernel trick (intuition)Planned
Part V · Evaluating & Improving Models
The skills that separate someone who calls .fit() from someone who can be trusted with a model.
Model Evaluation
Methodology- 19Train/validation/test, cross-validation & data leakagePlanned
- 20Metrics — accuracy, precision/recall, F1, ROC/AUC, regression metricsPlanned
- 21Hyperparameter tuning & model selectionPlanned
Part VI · Ensemble Methods
Why many weak models beat one clever model — the ideas behind the algorithms that win competitions on tabular data.
Ensembles
Supervised- 22Bagging & Random ForestsPlanned
- 23Boosting — AdaBoost, Gradient Boosting, XGBoost intuitionPlanned
Part VII · Unsupervised Learning
Finding structure when there are no labels — clustering and dimensionality reduction.
Unsupervised
No labels- 24k-Means clusteringFrom scratch
- 25Hierarchical clustering & DBSCANPlanned
- 26Dimensionality reduction & PCAPlanned
Part VIII · Putting It Together
End-to-end practice, and an honest look at the boundary where classical ML hands off to Deep Learning.
Practical ML & Capstone
Integration- 27Feature engineering & Scikit-learn pipelinesPlanned
- 28A full end-to-end project, start to finishPlanned
- 29Where classical ML ends & Deep Learning begins — and how to continuePlanned
Scope: deliberately stopping at Machine Learning
Per your goal, this path ends at classical ML. Deep Learning, neural networks, computer vision, NLP, reinforcement learning, and generative AI/LLMs are mentioned only to place ML in the wider landscape (Lessons 1, 3, and 29). Your mathematical foundations are handled separately at university — here, math appears only when it clarifies an idea.
Next step
New here? Start with Lesson 1 — What Is Artificial Intelligence?. Everything else is built on the mental map it gives you.
Already working through it? The most recent lesson is Lesson 12 — Multiple & Polynomial Regression: we widen the model to many features (the vectorized machinery fits them unchanged), make feature scaling a first-class step, and bend the line into curves with polynomial features — still linear in their parameters — all from scratch in NumPy. Next, Lesson 13 tackles overfitting, the bias–variance trade-off, and regularization.