Lectures
Lecture decks are designed for projection, keyboard navigation, and live stylus annotation. Detailed derivations and extended discussion belong in the textbook.
Available lectures
01 · Introduction to machine learning
Experience, tasks, performance measures, target scope, and the probabilistic roots of mean squared error.
02 · Logistic regression
Instructor-supplied slides covering exploratory data analysis, logistic regression, classification metrics, model settings, and likelihood-based losses.
03 · Decision trees and ensembles
The lecture covers decision trees; the accompanying tutorial covers ensembles. Both presentations are available as PDFs in Russian.
04 · Nonlinear representation learning with feedforward neural networks
Learned representations, artificial neurons, activation functions, and multilayer architecture.
05 · Visual representation learning with convolutional neural networks
Local visual evidence, convolutional feature hierarchies, deep residual architectures, normalization, transfer learning, and evidence-based model choice.
06 · Sequential representation learning with RNNs and Transformer encoders
Complete lecture for two 90-minute sessions: sequence problems, RNN forward/backward passes, gated memory, detailed attention and encoder calculations, five classroom questions, and a historical ending featuring Yoshua Bengio.
07 · Frequentist and Bayesian approaches to ML
The syllabus uses the Bayesian foundations and conceptual VAE introduction in Week 07. VAE loss/ELBO material in the existing deck is studied in Week 10; practical uncertainty estimation is scheduled for Week 12.
Trace RNN gradients, compare frequentist and Bayesian inference, derive MSE/BCE/CE from likelihood, explain L2 through MAP, and connect KL and Bayes to the VAE and ELBO. Includes visual examples, measured MNIST galleries, a Kingma spotlight, and optional practice with worked answers.
08 · Midterm task 30 review
The existing worked solutions for task 30 in the four released midterm variants: attention symmetry, ensemble variance, the ELBO and latent information.
09 · Time series and machine learning
Two 90-minute lectures: sampling and signal representations, Fourier and wavelets, causal features and validation, temporal neural models, EEGNet, Chronos-2 and a real SKAB pump case study. Includes interactive scenes, sliders, prediction checks and PDF key frames.
10 · Discovering Structure with Unsupervised Learning
PCA, clustering, autoencoders, t-SNE and latent generative models, with computed visualizations and VAE experiments.
During a lecture
- Use the arrow keys or swipe to move between slides.
- For Reveal.js decks, open the chalkboard controls to annotate.
- For Reveal.js decks, add
?print-pdfto the rendered slide URL for PDF printing. - For Week 9, use the PDF download links on the lecture page.