Lectures

Slides & recordings

Labs

• Lab1: linear regression with MSLE, cross-validation
• Lab2: backpropagation (presentation and leaflet on backpropagation)
• Lab3: MNIST again (presentation on softmax, NLL, CE)
• Lab4: PyTorch
• Lab5:
  • goldfish-to-shark (adversarial attack)
  • occlusion (saliency map)
  • dataset exploration
• Lab6: BatchNorm and ConvNets
• Lab7: ResNet and UNet
• Lab8: GAN
• Lab9: VAE (presentation)
• Lab10: Transformer and RNN
• Lab11: DQN
• Lab12: Policy Gradient
• Lab13: see Old exams below.
• Lab14 (Wed & Thu groups only): Consultations / more old exams.

Homeworks – themes & deadlines

• Moodle for homework submissions and grading

• HW1 (12.5 pt): Multitask Learning for Geometric Shape Classification and Counting, submissions, deadline Tuesday 25 XI, 23:59.
• HW2 (12.5 pt): GradCAM and SAM, submissions, deadline Tuesday 16 XII, 23:59.
• HW3 (15 pt): GPT-OSS-20B Transformer Architecture, submissions, deadline Tuesday, 13 I, 23:59.
• HW4 (10 pt): DQN Exploration, submissions, deadline Friday, 30 I, 23:59. Early deadline (for those aiming for exam pass): Tuesday, 27 I, 23:59.

Exam

The exam (both the quiz and programming session parts) will take place on:

• first date: Saturday, February 7th, 10:00-15:00, in the labs (rooms 2041–3045).
• retake: Thursday, February 19th, 9:00-14:00 15:00-20:30, in the labs (rooms 2042–2044).

Old exams

• 2020-21
• 2021-22
• 2022-23
• 2023-24
• 2024-25
• Quizzes (theory parts) 2020/21 – 2023/24

---

Grading rules

• 50 points for homework,
• 50 points for the exam,
• up to 10 points for activity – this is (in total) equivalent to ~+0.5 grade on the final course grade. These are awarded by lab instructors for extra effort (e.g. solving optional tasks, presenting solved scenarios from previous classes, finding a significantly better solution), up to 2.5 points at a time.

Total: 110 points possible. Grading scale:

• ≥ 90 points : 5.0
• ≥ 80 points: 4.5
• ≥ 70 points: 4.0
• ≥ 60 points: 3.5
• ≥ 50 points: 3.0

To pass, you need to pass both the homework and exam parts (≥50% for the homework and ≥50% for the exam as a whole). Lower thresholds for grades and for passing may be announced after grading is done.

Homework rules

There will be four homework assignments, in the form of larger lab-like scenarios to be solved individually at home. Each assignment will have a deadline at least one week after publication (at least 2 weeks for more complex tasks). There may be an uneven distribution of points for the four assignments.

Homework may be submitted within its announced deadline and/or late (until the start of the exam session, so by January 23rd, unless a given homework assignment’s graders allow later submissions).

Late submissions have their score multiplied by 0.6, even if you are late by 10 minutes. (If you are late by less than 10 minutes your score will be multiplied by 0.9.)

Exam rules

The exam will have two parts:

• Theoretical: quiz with multiple-choice and open questions. No materials allowed.
• Programming session. Materials allowed: code written earlier by yourself; internet for documentation and definitions only. Using code assistants or looking for solutions on the internet is NOT allowed.

Exam pass: there will be a free exam pass (with the highest grade) given to some number of students with the top total scores for homeworks (excluding points for lab activity). This requires submitting the last homework earlier.

---

ML Bootcamp – Introduction to Machine Learning

(open to all students interested in practical ML foundations and with any questions regarding DNN laboratories)

The bootcamp will take place during the first 4 weeks of the semester, on Mondays and Wednesdays at 18:15 (6:15 PM) in MiM 3043/44/45 building. Dates: 6.10, 8.10, 13.10, 15.10, 20.10, 22.10, 27.10, 29.10 Instructors: Michał Krutul, Jan Małaśnicki, Maciej Stefaniak

• Lab1&2: NumPy & Pandas basics

• Lab3: Linear Regression

• Lab4: Logistic Regression

• Lab4(extra): Softmax Regression

• Lab5: Experiment Tracking