Labs
Labs will assess experimental reasoning: problem formulation, validation design, comparison, error analysis, reproducibility, and interpretation.
Released labs
Week 1 — Linear regression
Train the same regression model with NumPy, native PyTorch autograd, and PyTorch Dataset/DataLoader; compare gradient calculations, learning progress, and validation results.
Week 2 — Logistic regression
Move from training-data EDA to four PyTorch logistic-regression pipelines; compare scaling, PCA compression, and a nonlinear quadratic representation while keeping validation and test roles separate.
Week 3 — Decision trees and ensembles
Enter the Hidden Pairs competition and build a baseline using decision trees and ensemble methods.
Week 5 — Satellite scene classification
Classify aerial and satellite images, complete the individual MLP/CNN checkpoint, then continue in the Kaggle competition.
Week 9 — SKAB: when should the pump stay in alarm?
Compare robust statistics, PCA, Isolation Forest, score fusion and recorded Chronos-2-small forecasts on pump measurements. The 90-minute student lab runs on CPU. Detect the whole faulty period, compare matched persistence residuals, and freeze a validation choice before examining held-out runs.
Open the SKAB lab and notebooks
| Language | Notebook | Google Colab |
|---|---|---|
| English | Download EN | Open EN in Colab |
| Русский | Скачать RU | Открыть RU в Colab |
If Colab cannot open a link, download the notebook and use File → Upload notebook in Colab. Save your own copy before editing.
Planned labs
Week 7 — Uncertainty estimation
CIFAR-10N lab space with a short introduction; assignment instructions will be added by the instructor.
Week 11 — Learn useful representations with limited labels
Week 13 — PINN or iPINN competition
Week 15 — Short final-week lab or tiny project
TODO(instructor): Publish the remaining assignment instructions, release details and competition requirements.