Deep Generative Models

This repository contains materials for the Deep Generative Models course taught at the Faculty of Computer Science of HSE University and Yandex School of Data Analysis.

Lectures: Artem Ryzhikov
Seminars: Aleksandr Khizhik
Assistants: Mariia Rubanenko

Grading

The final grade is calculated using the following formula:

$$ O_{\text{final}} = 8 \cdot 0.08 \cdot O_{\text{homework}} + 0.16 \cdot O_{\text{project}} + 0.2 \cdot O_{\text{exam}} $$

Where:

• Homework consists of 8 assignments with applied problems. Deadlines are strict (2 weeks) with no extensions allowed.
• Bonus points can be earned in the homework section by completing additional tasks.
• Project. See the Project Guidelines for details.
• Exam is conducted orally, using an open list of questions, with no additional preparation time provided.

Make sure to meet the deadlines and requirements for each component to achieve the best possible grade.

Project Guidelines

The project can take one of two forms:

1. Research Project: This involves reproducing code and results from recent academic papers. Only papers published in 2023, 2024, or 2025 are eligible. If multiple teams work on the same paper, they must differentiate their objectives (e.g., one team focuses on reproducing the results, while another explores modifications and improvements).

2. Startup Project: This involves developing a product based on existing solutions, typically in the form of a Telegram bot or another accessible application. The key requirement is that the product must be novel or have limited market availability. Any paper or technology can be used as a foundation, provided that the end result is sufficiently innovative.

Team Composition and Contributions

• Each team can have a maximum of four members.
• Every participant must demonstrate and justify their unique contribution during the final defense.

Open-Source Considerations

• If using open-source code, the project must include research beyond mere replication. Contributions should go beyond reproducing existing results by introducing enhancements, modifications, or novel interpretations.

The project should be well-documented and structured to ensure clarity and reproducibility.

Project Defense

The final defense will consist of the following components:

1. Team Presentation: Each team will deliver a structured presentation of their project, covering the motivation, methodology, key results, and conclusions. The presentation should clearly highlight each team member's contribution. The approximate time for the defense is 10-15 minutes per team.
2. Repository Submission: The team must submit a well-documented GitHub repository containing:
   • Source code.
   • A detailed README with instructions on reproducing results.
   • Any datasets or scripts required to run the project.
3. Working Solution (for Startup Projects): If the project is a startup-style application, the team must demonstrate a working prototype or deployment of the solution.
4. Q&A Session: The evaluation panel will ask questions regarding the project's implementation, results, and individual contributions.

Teams should ensure their documentation is clear and that all results are reproducible.