Free lecture notes from open courseware

University lectures published under an open licence are some of the best study material there is, and among the hardest to study from: the material is an hour of video with no table of contents. These pages are the reading version — every lecture broken into chapters with timestamps, its arguments pulled out as key points, and its vocabulary defined.

Everything here is free to read, with no account and no paywall. The collection is small and deliberately so: 7 lectures from 2 courses at the moment, added in small batches rather than bulk-published, because a page nobody can find is worth less than one that is actually read.

What each set of notes contains

• A chapter breakdown with timestamps — the lecture divided into its sections, each with a short summary and the time it starts, so you can jump to the part you need in the original video.
• Key points — the claims and results the lecture actually argues for, as single lines.
• Key terms — the vocabulary introduced, defined in the sense the lecturer used it rather than a dictionary sense.
• A link to the source lecture — every page names the course, the term and the lecturer's institution, and links the video.

MIT 6.006 Introduction to Algorithms

Fall 2020, published openly by MIT OpenCourseWare. 1 of its lectures have notes here; the chapter titles under each one are the sections the notes actually contain, so you can tell before opening a page whether it covers what you came for.

• Lecture 1: Algorithmic Thinking, Peak Finding 3 chapters · 36 key terms

  Covers: Course Overview & Core Topics · One‑Dimensional Peak Finding · Two‑Dimensional Peak Finding

MIT 6.7960 Deep Learning

Fall 2024, published openly by MIT OpenCourseWare. 6 of its lectures have notes here; the chapter titles under each one are the sections the notes actually contain, so you can tell before opening a page whether it covers what you came for.

• Lec 01. Introduction to Deep Learning 10 chapters · 77 key terms

  Covers: Deep Learning Course Overview · Course Overview & Policies · Historical Peaks of Neural Networks · Linear Layers & Activation Functions · ReLU and Layer Stacking · Non‑linear Approximation & Generalization · Data, Model Size, and Loss · Gradient, Batching & Representations · Scaling & Hype in Deep Learning

• Lec 02. How to Train a Neural Net 7 chapters · 84 key terms

  Covers: Opening Overview · Training Neural Networks Basics · Computational Graphs & Backpropagation · Custom Operations in PyTorch · Closing Remarks

• Lec 03. Approximation Theory 12 chapters · 67 key terms

  Covers: Universal Approximation & Depth vs Width · Three‑Layer ReLU Approximation Bounds · Rectangle‑Based Function Approximation · Approximating Functions with ReLU Rectangles · Constructing Hyper‑Rectangles with ReLU Networks · Universal Approximation & Width vs Depth · Depth Separation and Width · ReLU Layers and Kink Growth · Adapting Transformers to Data Structure

• Lec 04. Architectures: Grids 12 chapters · 73 key terms

  Covers: MLP Foundations & Limitations · Model Choice & Generalization · Patch-based Segmentation · Convolution Basics · Convolution Fundamentals · Convolutional Layer Mechanics · CNN Building Blocks · Stride, Filters, and Receptive Fields · ResNet & Temporal Convolutions · Neural Fields & Positional Encodings · NeRF & Video Representation

• Lec 05. Architectures: Graphs 12 chapters · 75 key terms

  Covers: Graph Neural Networks Overview · Graphs in Real‑World Systems · Graph Neural Network Fundamentals · Local Operations in CNNs and GNNs · Aggregate Functions in GNNs · Graph Neural Networks Basics · GNN Expressiveness Limits · GNN Expressiveness & Aggregation · Breaking GNN Limits

• Lec 06. Generalization Theory 11 chapters · 67 key terms

  Covers: Generalization in Deep Learning · Generalization vs Memorization · Model Memorization vs. Generalization · From Overfit to Double Descent · Neural Nets vs Polynomial Regression · Inductive Biases & Version Space · Bias Toward Simplicity · Depth, Bias, and Symmetry

What this collection is not

• Not transcripts. No page reproduces a lecture verbatim. If you need the exact words, the source video is linked on every page and carries its own captions.
• Not the course. There are no problem sets, no solutions and no exams here, and notes are no substitute for watching a lecture you are being examined on.
• Not affiliated with MIT. The lectures are MIT OpenCourseWare's and openly licensed; the notes are ours, and the institution has neither reviewed nor endorsed them.
• Not error-free. These are machine-generated from audio. Chapter titles and definitions are usually right and occasionally are not, which is the honest trade for notes existing at all.