Knowledge Discovery and Pattern Extraction

About This Course

As AI automates more of our work, one skill remains irreplaceable: the ability to understand and evaluate how systems behave at scale. No matter how advanced AI becomes, humans will always be asked to judge whether results can be trusted, whether measurements are meaningful, and whether decisions are sound. This is the final frontier.

This course teaches the science of extracting reliable knowledge from data. You will learn to design rigorous experiments, quantify uncertainty, avoid common pitfalls in evaluation, and apply techniques for similarity search, clustering, and graph analysis.

The emphasis is on doing it right: understanding when measurements can be trusted, how bias and variance corrupt decisions, and how to scope and execute data analysis projects that deliver actionable insights.

Learning Objectives

By the end of this course, students will develop five core competencies:

1. Making Sense of Data at Scale

• Apply similarity search, clustering, and LSH to find patterns in large datasets
• Analyze graph-structured data: compute centrality, detect communities, reason about network structure
• Scope and execute data analysis projects from problem definition to actionable insights

2. Understanding and Estimating the Quality of AI Systems

• Design rigorous evaluations: train/val/test splits, avoid data leakage, choose appropriate metrics
• Reason about bias and variance: identify systematic vs random error, understand how selection amplifies optimism
• Distinguish correlation from causation: recognize confounding, Simpson's paradox, and other pitfalls

3. Designing Experiments and Communicating Results

• Design experiments that answer specific questions
• Communicate results that convey: (a) what we wanted to learn, (b) what we learned, and (c) what remains uncertain
• Quantify and report uncertainty in findings

"A measurement without knowledge of its uncertainty is meaningless." — Walter Lewin

4. Critical Evaluation of Data-Driven Claims

• Separate what the data tell us from what they don't
• Identify when conclusions overreach the evidence
• Ask the right questions about methodology, sampling, and alternative explanations
• Recognize common patterns of overstated certainty

5. Creating Representations that Capture Complex Systems

• Use embeddings to represent text, users, and items in vector spaces
• Build graph representations that capture relationships and structure
• Combine embeddings and graphs to extract knowledge from complex systems

Sessions

Part 1 · Foundations

Part 2 · The Science of Experimentation and Uncertainty

Part 3 · Representations and Abstractions

Part 4 · Mastering Graphs

Part 5 · Conversational "X"

Prerequisites

• Linear algebra
• Python programming
• Familiarity with classification and regression

Assessment

See full exam details, rubrics, and policies →

Policies

Academic Integrity

Exams are individual work. Projects are individual. You may use AI tools for learning but must understand and explain anything you submit.

AI Tools

You may use LLMs (ChatGPT, Copilot, etc.) for coding assistance and exploration. Document AI usage in your submissions.