The Forge · engineering, computing & technology
AI & Machine Learning
What it means for a machine to learn — the mathematics, the code, and the limits.
Regression to random forests — fitting models, judging them honestly, and knowing when they lie.
Syllabus · 4 units · ~40 hours
Unit I — Learning as Fitting
The supervised learning setup · Linear regression and least squares · Loss functions · Train, validate, test
Unit II — Classification
Logistic regression · Decision trees · k-nearest neighbors · Precision, recall, and the confusion matrix
Unit III — Overfitting and Its Cures
Bias and variance · Regularization · Cross-validation
Unit IV — Ensembles and Practice
Random forests · Gradient boosting · Feature engineering · A full project on a real dataset
From a single neuron to a trained deep network — backpropagation worked by hand, then by machine.
Syllabus · 4 units · ~44 hours
Unit I — The Neuron and the Layer
Perceptrons · Activation functions · Why depth helps
Unit II — Training
Gradient descent · Backpropagation, step by step · Learning rates and optimizers · Batch normalization
Unit III — Architectures
Convolutional networks for images · Recurrent networks and sequences · Attention and transformers
Unit IV — Practice and Pitfalls
Data augmentation · Transfer learning · Reading a loss curve · When the model memorizes instead of learning
Tokens, embeddings, transformers — what large language models actually do, without the fog.
Syllabus · 4 units · ~30 hours
Unit I — Words as Numbers
Tokenization · Embeddings and similarity · n-gram models and their ceiling
Unit II — The Transformer
Attention, plainly explained · Context windows · Pretraining and next-token prediction
Unit III — From Model to Assistant
Fine-tuning and instruction following · Prompting as an interface · Hallucination and grounding
Unit IV — Judgment
Evaluating model output · Bias in training data · What these systems cannot do
Pixels to predictions — filters, convolutions, and the networks that read images.
Syllabus · 4 units · ~36 hours
Unit I — The Image as Data
Pixels, channels, and color spaces · Filters and convolution by hand · Edges and gradients
Unit II — Classical Vision
Feature detection · Image transforms · Template matching and its limits
Unit III — Deep Vision
Convolutional networks revisited · Object detection · Segmentation
Unit IV — Vision in the Field
Datasets and labeling · Evaluation metrics · Failure modes and adversarial examples
Who a model serves, who it fails, and how to reason about systems no one fully controls.
Syllabus · 3 units · ~18 hours
Unit I — Where Bias Comes From
Training data as a mirror · Measurement and proxy variables · Case studies in unfair models
Unit II — Accountability
Explainability and its limits · Auditing a model · Regulation in outline
Unit III — The Wider Ledger
Labor and automation · Energy and compute costs · What to demand from systems that decide