How to Study using Obsidian

Studying effectively in Obsidian hinges on splitting note-taking into two phases: fast, context-based capture during learning, then deliberate, connected “long-term knowledge management” afterward. The payoff is a living knowledge base where key concepts evolve over time and link back to the original contexts that shaped them—so understanding doesn’t stay trapped inside one lecture or one folder.

In the first phase, notes look familiar: a course folder with subfolders by week, and within each week, notes for each lecture. While watching “Introduction to Machine Learning,” the workflow mirrors typical study behavior—write down definitions, formulas, and explanations that seem salient in the moment. For example, while covering regression models, the notes include supervised learning framing, training data, the mathematical form for linear regression, cost functions (including what they mean with diagrams), and gradient descent. Markdown structure—subheadings and sections—helps the notes stay readable and searchable later, but the emphasis remains on capturing what matters during the learning session.

The second phase is where Obsidian’s connected-note approach becomes central. After a class or at the end of a week, the notes get reviewed with a new goal: identify the key concepts worth turning into durable “atomic” notes, then link them together like a personal Wikipedia. Instead of leaving “training data” as a one-off definition inside a lecture note, the workflow creates a dedicated note for training data and back-links it to where the idea first appeared. Even if this feels duplicative at the start, it enables future updates: when later lectures reference training data in a new way, the atomic note can be expanded or refined, while the backlinks preserve the original context.

This curation-and-linking process also forces deeper understanding. When the cost function section grows dense, the workflow extracts a clear definition in the learner’s own words, then links it back to the lecture material. For gradient descent, the workflow demonstrates a “live” state: some parts are curated and defined, while other items remain placeholders for later refinement—such as noting that other gradient descent variants exist even if they weren’t covered yet. Crucially, the notes can link to specific sections within source notes, not just entire documents.

As the course progresses, the network starts to pay off. Gradient descent reappears in “multiple linear regression,” bringing new details like the normal equation as an alternative to gradient descent and practical guidance on convergence checks via the learning curve. Those additions get folded back into the gradient descent atomic note, so knowledge accumulates in one place rather than fragmenting across separate lecture notes. The result is a continuously updated concept map: local graphs summarize what was learned in a course (e.g., training data, cost function, linear regression, gradient descent) and show how related concepts connect (e.g., learning curve, batch gradient descent, normal equation). The central claim is practical: connected note-taking turns study notes into a system that supports retrieval, revision, and long-term understanding across any subject—not just data science.