Skip to content

Handouts

Full class walkthroughs as single-page documents. Each one renders every figure from real training runs, ships the code to reproduce it, and includes solved checkpoints.

They complement the class pages rather than replace them: each goes deep on one topic and works both as a companion during class and as a reference afterwards.

Written in Portuguese

The handouts are course material for the in-person classes and are written in Brazilian Portuguese. The code, the library names and the figures are language-independent.

  • Training an MLP: batch, optimizer and learning rate


    MNIST — 70 000 handwritten digits, one fixed architecture: 784 → 128 → 64 → 10.

    The three decisions that are neither the architecture nor the update rule. Full-batch, mini-batch and pure SGD measured against each other by epoch, by weight update and by wall clock; gradient noise measured directly against the 1/√B prediction; the seven optimizers compared first at one shared learning rate and then each at its own; a 6 × 6 batch × learning-rate map and a 7 × 9 optimizer × learning-rate grid, both explorable in the page; and what a loss curve has to look like before it means anything.

    ~2 h · 10 figures · 2 live simulators · 7 checkpoints with solutions

    open · colab

Classroom use

Each handout has an anchored sidebar index, a copy button on every code block, and its own print stylesheet — the menu, buttons and interactive elements are dropped when printing or exporting to PDF.