CENG0056 Deep Generative Models

6 ECTS - 3-0 Duration (T+A)- . Semester- 3 National Credit

Information

Code CENG0056
Name Deep Generative Models
Term 2024-2025 Academic Year
Semester . Semester
Duration (T+A) 3-0 (T-A) (17 Week)
ECTS 6 ECTS
National Credit 3 National Credit
Teaching Language Türkçe
Level Yüksek Lisans Dersi
Type Normal
Mode of study Yüz Yüze Öğretim
Catalog Information Coordinator Mehmet SARIGÜL


Course Goal / Objective

The goal of a Deep Generative Models course is to provide students with a comprehensive understanding of generative modeling techniques using deep learning architectures. Deep generative models aim to learn and generate new data samples that resemble a given dataset, capturing its underlying distribution and structure.

Course Content

This course covers the Introduction to Generative Modeling, Overview of generative modeling, Probabilistic modeling and likelihood estimation, Maximum Likelihood Estimation (MLE), Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Autoregressive Models, Deep Generative Models in Natural Language Processing, Advanced Topics in Deep Generative Models, Applications of Deep Generative Models, Critiquing Research Papers.

Course Precondition

Knowledge of basic programming, linear algebra, and probability theory.

Resources

Tomczak, J. M. (2022). Deep generative modeling (pp. 1-197). Springer.

Notes

Tomczak, J. M. (2022). Deep generative modeling (pp. 1-197). Springer.


Course Learning Outcomes

Order Course Learning Outcomes
LO01 Understanding of Generative Modeling
LO02 Familiarity with Deep Learning Architectures
LO03 Ability to Train and Evaluate Deep Generative Models
LO04 Application of Deep Generative Models


Relation with Program Learning Outcome

Order Type Program Learning Outcomes Level
PLO01 Bilgi - Kuramsal, Olgusal On the basis of the competencies gained at the undergraduate level, it has an advanced level of knowledge and understanding that provides the basis for original studies in the field of Computer Engineering. 4
PLO02 Bilgi - Kuramsal, Olgusal By reaching scientific knowledge in the field of engineering, he/she reaches the knowledge in depth and depth, evaluates, interprets and applies the information. 3
PLO03 Yetkinlikler - Öğrenme Yetkinliği Being aware of the new and developing practices of his / her profession and examining and learning when necessary. 3
PLO04 Yetkinlikler - Öğrenme Yetkinliği Constructs engineering problems, develops methods to solve them and applies innovative methods in solutions. 2
PLO05 Yetkinlikler - Öğrenme Yetkinliği Designs and applies analytical, modeling and experimental based researches, analyzes and interprets complex situations encountered in this process. 3
PLO06 Yetkinlikler - Öğrenme Yetkinliği Develops new and / or original ideas and methods, develops innovative solutions in system, part or process design. 3
PLO07 Beceriler - Bilişsel, Uygulamalı Has the skills of learning. 2
PLO08 Beceriler - Bilişsel, Uygulamalı Being aware of new and emerging applications of Computer Engineering examines and learns them if necessary. 2
PLO09 Beceriler - Bilişsel, Uygulamalı Transmits the processes and results of their studies in written or oral form in the national and international environments outside or outside the field of Computer Engineering.
PLO10 Beceriler - Bilişsel, Uygulamalı Has comprehensive knowledge about current techniques and methods and their limitations in Computer Engineering. 2
PLO11 Beceriler - Bilişsel, Uygulamalı Uses information and communication technologies at an advanced level interactively with computer software required by Computer Engineering. 2
PLO12 Bilgi - Kuramsal, Olgusal Observes social, scientific and ethical values in all professional activities. 2


Week Plan

Week Topic Preparation Methods
1 Introduction to generative modeling, probabilistic modeling, and likelihood estimation. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
2 Autoencoders and their limitations, introduction to Variational Autoencoders (VAEs). Reading the lecture notes Öğretim Yöntemleri:
Anlatım
3 Variational inference, evidence lower bound (ELBO), and training VAEs. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
4 Evaluation of VAEs, sampling from the latent space, and reconstruction quality. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
5 Introduction to Generative Adversarial Networks (GANs) and the generator-discriminator framework. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
6 GAN training and the GAN objective, variations of GANs (e.g., DCGAN, WGAN). Reading the lecture notes Öğretim Yöntemleri:
Anlatım
7 Challenges in GAN training (mode collapse, instability) and solutions. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
8 Mid-Term Exam Ölçme Yöntemleri:
Yazılı Sınav
9 Autoregressive models, PixelCNN, and PixelRNN. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
10 Normalizing Flows and flow-based generative models. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
11 Language modeling with deep generative models, text generation using RNNs and transformers. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
12 Conditional generation and text-to-image synthesis. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
13 Disentangled representation learning and its applications. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
14 Diffusion Models. Reading the lecture notes Öğretim Yöntemleri:
Anlatım
15 Review Reading the lecture notes Öğretim Yöntemleri:
Anlatım
16 Term Exams Ölçme Yöntemleri:
Yazılı Sınav
17 Term Exams Ölçme Yöntemleri:
Yazılı Sınav


Student Workload - ECTS

Works Number Time (Hour) Workload (Hour)
Course Related Works
Class Time (Exam weeks are excluded) 14 3 42
Out of Class Study (Preliminary Work, Practice) 14 5 70
Assesment Related Works
Homeworks, Projects, Others 0 0 0
Mid-term Exams (Written, Oral, etc.) 1 14 14
Final Exam 1 28 28
Total Workload (Hour) 154
Total Workload / 25 (h) 6,16
ECTS 6 ECTS

Update Time: 13.05.2024 03:07