Machine Learning Essentials
You've heard that machines learn from data. But how, exactly? This course takes you inside the black box. Not with code — with understanding. You'll learn to think like a machine learning practitioner: recognising which problems suit machine learning, preparing data so models can actually learn from it, and understanding the algorithms that power everything from spam filters to self-driving cars. You'll start with the thinking behind it — how to look at a real-world question and turn it into something a machine can learn to answer. Then you'll see how data needs to be shaped before a model can use it, and the subtle mistakes that make systems look brilliant in testing and fail in the real world. From there, you'll meet the core algorithms — the different approaches machines use to find patterns, make predictions, and group things together. Some are simple enough to follow by hand. Others combine hundreds of smaller models into something far more powerful. You'll understand when each approach makes sense and what tradeoffs come with it. You'll learn how models improve — how they measure their own mistakes and gradually get better, and why there's a constant tension between learning the training data too well and not learning it well enough. You'll see why measuring success is harder than it sounds, and why the obvious metric is often the wrong one. You'll go deeper into neural networks — the technology behind image recognition, language models, and most of modern AI. You'll understand how they actually work, why different designs suit different kinds of data, and why almost nobody builds them from scratch anymore. The course closes with what happens after a model is built: how ML systems work in production, how they degrade over time, and the practices that make them not just accurate but fair and trustworthy. No programming required. But after this course, you'll understand what the programmers are actually doing.
Section 1: The ML Mindset
Machine learning isn't just a set of algorithms — it's a way of thinking about problems. Before you touch any data, you need to recognise which questions can be answered by learning from examples, which are better solved by writing rules, and which shouldn't be automated at all. You need to understand the tradeoff between getting the most accurate answer and getting one you can explain. And you need to see the end-to-end process that every ML project follows, from defining the question to deploying something that works. This section gives you the practitioner's mindset — the thinking that comes before the doing.
Section 2: Preparing the Ground
A model can only learn from what you give it. Raw data — messy, incomplete, full of irrelevant noise — needs to be shaped into something a model can work with. The inputs need to be chosen and transformed. The answers need to be defined. And the data needs to be split carefully so you can test your model honestly, without fooling yourself. This section is about the work that comes before the algorithm — the work that determines whether the algorithm has any chance of succeeding.
Section 3: Core Algorithms — Prediction and Explanation
It's time to meet the machines that learn. This section introduces two foundational algorithms that every ML practitioner should understand deeply. The first draws a line through data to make predictions — and despite its simplicity, it teaches you nearly everything about how models work. The second asks a series of yes/no questions to reach a decision, producing a result that a human can read and follow like a flowchart. Together, they represent two poles of machine learning: pure prediction and transparent explanation. Understanding both — and the tension between them — is essential.
Section 4: Core Algorithms — Combining and Grouping
One model can be wrong in unpredictable ways. But if you build many models — each seeing the data slightly differently — and let them vote, the errors cancel out and what remains is remarkably accurate. This section covers the ensemble methods that dominate real-world machine learning, and the approaches that find structure in data without being told what to look for.
Section 5: How Models Learn
Every model needs two things: a way to know it's wrong, and a method to get less wrong. The first is a function that measures the gap between what the model predicted and what actually happened. The second is a process that adjusts the model, step by step, to close that gap. But there's a fundamental tension — a model that fits the training data perfectly will often fail on new data. Too much learning is as dangerous as too little. This section is about the mechanics underneath every algorithm you've met so far.
Section 6: Measuring Success
How do you know if your model works? The obvious answer — count how often it's right — is often dangerously misleading. A model that predicts "no disease" for every patient could be 99% accurate if only 1% of patients are sick. It would also be completely useless. Measuring ML models properly means asking the right questions about the right kinds of mistakes, getting performance estimates you can trust, and understanding the theoretical tension at the heart of why models fail.
Section 7: Neural Networks Demystified
If you completed AI Fundamentals, you know what neural networks are and why they matter. Now you'll understand how they actually work. A single artificial neuron turns out to be something you already understand — just a simple mathematical function. Stack many of them in layers and you get a system that can learn extraordinarily complex patterns. Different arrangements suit different kinds of data — some are designed for images, others for language. And thanks to a practice called transfer learning, almost nobody builds these systems from scratch anymore.
Section 8: ML in the Real World
A model that works in an experiment is the beginning, not the end. In the real world, data flows through pipelines, experiments need to be tracked and reproduced, and models quietly degrade as the world changes around them. And increasingly, accuracy isn't enough — you need to be able to explain what your model does and demonstrate that it treats people fairly. This section is about the gap between building a model and running one responsibly.