Math foundations
Start with the calculation you want to understand. Each sequence builds the notation around worked numbers, with extra detail available when you need it.
| Your question | Start here |
|---|---|
| How does a table of inputs become a prediction? | Linear algebra: vectors, matrix multiplication, and least squares |
| How quickly does a quantity change, and what do those changes add up to? | Calculus: derivatives, integrals, and gradients |
| How do I describe uncertainty and calculate an average? | Probability: distributions, expectation, and conditioning |
| How does a model’s error tell us which parameters to change? | Partial derivatives and backpropagation: one loss, its derivatives, and a backward pass |
From calculus to training
Section titled “From calculus to training”Start with derivatives if rates of change are unfamiliar. Then Backpropagation follows one small prediction through its loss and parameter updates. Its optional Jacobian reference explains the matrix notation used in lectures.
The math beneath AI looks at logarithms, odds, entropy, and cross-entropy. Use it to investigate a particular operation you encounter in the Transformer sequence.
Put the ideas to work
Section titled “Put the ideas to work”Learning from data uses these tools to fit and evaluate predictions. One complete dataset connects averages, squared error, a gradient step, validation, and a final test.