Engineering / core
Build and understand learning systems
Learn how models acquire useful behavior, how training fails, and how to evaluate evidence.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 core
Build and understand learning systems
Learn how models acquire useful behavior, how training fails, and how to evaluate evidence.
- 01
Learning problems and trustworthy data
Turn a product question into a measurable prediction task without leaking the answer into the inputs.
24 min - 02
Linear regression as a complete learning experiment
Fit a transparent numerical model, derive its gradient, and diagnose what residuals reveal.
26 min - 03
Classification, probabilities, and decision costs
Connect logits, cross-entropy, and thresholds without confusing a confidence score with a decision.
27 min - 04
Backpropagation through a computation graph
Trace how a scalar loss sends sensitivity backward through shared operations and learn to verify a gradient.
29 min - 05
Training dynamics and optimization failure
Understand step size, stochastic gradients, and adaptive optimizers by investigating when an update fails.
30 min - 06
Generalization, regularization, and honest selection
Understand how fitting a finite sample differs from predicting new data, and how to investigate that gap.
28 min - 07
Attention and the Transformer computation
Compute a small attention example, understand causal masks, and separate architecture from capability claims.
34 min - 08
Generative models: factorization and denoising
Compare autoregressive and diffusion objectives through their assumptions, training signals, and sampling procedures.
34 min - 09
Evaluation, uncertainty, and calibrated confidence
Measure ranking, decisions, and probability quality separately, with attention to prevalence and sample size.
31 min - 10
Training systems and reproducible experiments
Make an experiment restartable and interpretable by controlling data order, state, reductions, and evaluation behavior.
32 min