That is not a disclaimer — it is part of the point. The slides and the voice narrating the lectures were produced with AI tools, but every script was written and fact-checked by a human instructor with real hands-on experience. Think of the AI as the production crew; the teaching is human. Here’s a sentence that reshapes how you think about a whole category of software: a machine-learning system is not programmed — it’s grown from data. Nobody wrote the rules it follows. So its quality is determined by its data, expressed through statistical performance, and prone to silently decaying as the world changes. Traditional testing has almost no vocabulary for this.
What you will learn:
Testing Machine Learning Systems gives you that vocabulary — and a rigorous method. You’ll test the data, which is effectively the program. You’ll understand the metrics that define quality — and recognize the accuracy paradox and the aggregates that hide catastrophic failures. You’ll slice performance to find where a model quietly fails, probe robustness and fairness, and confront drift and monitoring, because an ML model is never truly done. No data-science degree required — just clear, statistical thinking.
What this course covers, module by module:
What Makes ML Different; The ML Lifecycle; Testing the Data; Understanding Metrics; Metrics That Mislead; Performance & Slices; Testing Robustness; Fairness & Bias; Data & Concept Drift; Monitoring in Production; The ML Pipeline; Explainability & Debugging; Pitfalls & Responsible ML Testing.
