Deep Learning 2

Data is displayed for the academic year: 2024./2025.

Lecturers

Prof.
Marko Subašić
PhD
Prof.
Zvonko Kostanjčar
PhD

Laboratory exercises

Tomislav Kurtović
univ. mag. ing.
Mario Tropčić
univ. mag. ing. comp.
Lovre Mrčela
mag. ing.

Course Description

The idea of generating new but convincing samples has always been interesting in the fields of artificial intelligence and machine learning and is generally considered a more difficult problem than the problems solved by discriminative models. Recent advances in specialized deep model architectures, combined with advances in optimization methods, have enabled the successful modeling of complex, high-dimensional data including images, text, and speech. In this course, we will study mathematical foundations and learning algorithms for deep generative models, including popular families of generative models such as variational autoencoders, generative adversarial networks, autoregressive models, normalizing flow models, and energy function-based models. Through practical laboratory exercises, the design, training and testing of some of the most famous deep generative models will be experienced.

Study Programmes

University graduate
[FER3-HR] Audio Technologies and Electroacoustics - profile
Elective Courses (3. semester)
[FER3-HR] Communication and Space Technologies - profile
Elective Courses (3. semester)
[FER3-HR] Computational Modelling in Engineering - profile
Elective Courses (3. semester)
[FER3-HR] Computer Engineering - profile
Elective Courses (3. semester)
[FER3-HR] Computer Science - profile
Core-elective courses (3. semester)
[FER3-HR] Control Systems and Robotics - profile
Elective Courses (3. semester)
Elective Courses of the Profile (3. semester)
[FER3-HR] Data Science - profile
Elective Courses (3. semester)
Elective Courses of the Profile (3. semester)
[FER3-HR] Electrical Power Engineering - profile
Elective Courses (3. semester)
[FER3-HR] Electric Machines, Drives and Automation - profile
Elective Courses (3. semester)
[FER3-HR] Electronic and Computer Engineering - profile
Elective Courses (3. semester)
[FER3-HR] Electronics - profile
Elective Courses (3. semester)
[FER3-HR] Information and Communication Engineering - profile
Elective Courses (3. semester)
Elective Coursesof the Profile (3. semester)
[FER3-HR] Network Science - profile
Elective Courses (3. semester)
[FER3-HR] Software Engineering and Information Systems - profile
Elective Course of the profile (3. semester)
Elective Courses (3. semester)

Learning Outcomes

  1. Define main notions of deep generative models.
  2. Explain the underlying technology of deep generative models.
  3. Describe the differences between the most prominent deep generative models.
  4. Evaluate different methods for sampling from an unknown probability distribution.
  5. Identify scenarios where deep generative models can be applied.

Forms of Teaching

Lectures

The lectures present theoretical concepts and algorithms followed by concrete examples.

Laboratory

Laboratory exercises are organized as projects.

Grading Method

Continuous Assessment Exam
Type Threshold Percent of Grade Threshold Percent of Grade
Laboratory Exercises 50 % 20 % 50 % 20 %
Seminar/Project 20 % 20 % 20 % 20 %
Mid Term Exam: Written 20 % 30 % 0 %
Final Exam: Written 20 % 30 %
Exam: Written 50 % 60 %
Comment:

The threshold on the sum of the midterm and the final exam is 50%.

Week by Week Schedule

  1. Course overview, introduction to deep generative learning.
  2. Mathematical Foundation & Basic Concept for generative learning.
  3. Basic concepts of deep learning, deep models and unsupervised learning.
  4. Traditional approaches to generative modeling.
  5. Autoregressive Models.
  6. Classical autoencoders.
  7. Variational autoencoders.
  8. Midterm.
  9. Advanced variational autoencoders.
  10. Normalising flow models.
  11. Generative adversarial networks.
  12. Advanced generative adversarial networks.
  13. Energy based models.
  14. Deep energy based models.
  15. Final exam.

Literature

Ian Goodfellow, Yoshua Bengio, Aaron Courville (2016.), Deep Learning, MIT Press
Eli Stevens, Luca Antiga, and Thomas Viehmann (2020.), Deep Learning with PyTorch, Manning Publications

General

ID 230255
  Winter semester
5 ECTS
L0 English Level
L2 e-Learning
30 Lectures
0 Seminar
0 Exercises
15 Laboratory exercises
0 Project laboratory
0 Physical education excercises

Grading System

87 Excellent
75 Very Good
63 Good
51 Sufficient

Similar Courses