Week 9 · Time series and machine learning
Two lectures of 90 minutes each, followed by a separate 90-minute laboratory. Both slide decks are in English.
Lecture 1 · Listening to time
From sensor readings to useful features: sampling and aliasing, temporal properties, Fourier coefficients and harmonics, frequency units, time–frequency resolution, wavelets, causal features and validation splits. Follow one pump through interactive scenes with step controls and sliders.
Lecture 2 · What is the signal telling us?
Five prediction tasks, temporal neural architectures, EEG and EEGNet, and a detailed Chronos-2 forecasting walkthrough. The SKAB case study distinguishes forecast quality, detector scores and actual alarms, and compares complete pipelines without assuming which model family should win.
Explore the ideas
Interactive pump scenes · Learning guide
Use the slide navigation controls or arrow keys. In each interactive scene, advance one step at a time, then change a slider to test your explanation. The PDFs preserve selected frames for reading and printing.
Put the ideas to work
Open the SKAB laboratory — EN/RU student notebooks
The laboratory runs on free Colab CPU. It asks whether different representations, score calibration and detector combinations help identify the whole faulty period, with onset detection as a secondary question.