Worked Examples

Trace a two-layer classifier from logits through cross-entropy and gradients. Then trace one masked attention head, including shapes and mask values. Compare full precision with quantized inference and reject the compressed candidate if quality loss exceeds the frozen tolerance.

Evidence and review

Show calculations, code, measurements, and the rejected alternative. Claims must trace to versioned code and artifacts. Screenshots, unexplained generated prose, and a single favorable run do not pass.

Oral defense

Explain the mechanism, identify the most dangerous assumption, interpret one failure from raw evidence, and name the condition that would reverse your decision.

Source backbone

Use Dive into Deep Learning, PyTorch documentation, primary papers, and Designing Data-Intensive Applications selectively.