Living open educational resource · v1.0 · May 2026
Machine Learning by Design
From problem framing to reliable systems.
Machine Learning by Design: From Problem Framing to Reliable Systems is a concept-first open educational resource in machine learning for IOAI students, computer science undergraduates, instructors, and serious self-learners. Instead of presenting machine learning as only a catalog of models or a software toolkit, the book teaches it as a discipline of problem framing, evidence, evaluation, reliability, and system design.
The book covers task formulation, predictive modeling, loss functions, decision rules, evaluation, baselines, linear models, trees, neural networks, representation learning, foundation-model workflows, modality-specific modeling, experiments, uncertainty, reliability, and deployment-minded AI systems. Its central goal is to help readers move from a vague real-world problem to a defensible model claim and then to a reliable system.
The material is organized around a recurring set of practical questions: what decision is being supported, what learning task is the closest defensible proxy, what data stands in for the world, what representation exposes the relevant structure, what baseline is serious, what evidence justifies the claim, and how a model output becomes an action in a real system. The aim is to give readers both technical foundations and a durable way of thinking about building and evaluating machine learning systems.
About this edition
First stable release: v1.0, May 2026. This is a living open educational resource — maintained on a stated cadence with monthly minor revisions for errata and small additions, and larger updates aligned with the academic calendar in September and January. The book is freely available under CC BY-NC-SA 4.0. Revision history is tracked in the project's CHANGELOG.
Adopt in your course
The book is adoptable in undergraduate machine learning and AI courses today. The Instructor Guide includes 10-week and 12-week course paths, a chapter-dependency map, assignment patterns, and a four-part rubric. Sample instructor solutions are released alongside the book and expanded with each major update.
Cite this book
Zenodo DOI: 10.5281/zenodo.19341954
@book{urmian2026mlbd,
author = {Urmian, Sam},
title = {Machine Learning by Design: From
Problem Framing to Reliable Systems},
year = {2026},
edition = {First stable edition (v1.0)},
doi = {10.5281/zenodo.19341954},
url = {https://github.com/mlgorithm/ml-by-design},
note = {Open-access edition. CC BY-NC-SA 4.0.}
}Report errata
Found an error? Open a GitHub issue with the label errata at github.com/mlgorithm/ml-by-design/issues, or email the author. Confirmed errata are listed in ERRATA.md with the affected page, the correction, and the version that incorporates the fix.
Repository
Contact
Sam Urmian, SLATE — Centre for the Science of Learning & Technology, University of Bergen. sam.urmian@gmail.com