Differential Privacy in the Era of Foundation Models
Semester: Winter 2024/2025
Added: Oct 23, 2024
Winter 2024/2025
Instructor: Franziska Boenisch and Ben Stock
Contact: franziska.boenisch@cispa.de
Lectures: Wednesdays 16:10 - 18:00, CISPA building (Stuhlsatzenhaus 5, Saarbrücken)
Description
In recent years, foundation models like GPT, LLaMA, or Stable Diffusion have transformed AI. This seminar explores how differential privacy (DP) can be applied to foundation models to safeguard data privacy. Key topics include DP theory, practical mechanisms, and privacy-preserving architecture studies.
Assignments
Each student will present one or two topics during the seminar hours and submit a comprehensive seminar paper at the end of the semester.
Schedule
| Date | Topic | Location |
|---|---|---|
| 23.10.2024 | Introduction: Presentation of Seminar Topics, and “How-To” give a presentation | Room 0.02 |
| 30.10.2024 | Topic 1: Introduction to Foundation Models & The Pre-train/Adapt Paradigm | Room 0.02 |
| 13.11.2024 | Topic 2: Introduction to Differential Privacy | Room 0.02 |
| 20.11.2024 | Topic 3: Privacy Risks in Foundation Models | Room 0.02 |
| 18.12.2024 | Topic 4: Privately Pre-Training Diffusion Models | Room 0.02 |
| 08.01.2025 | Topic 5: Privately Fine-Tuning Diffusion Models | Room 0.02 |
| 15.01.2025 | Topic 6: Privately Training Large Language Models | Room 0.02 |
| 22.01.2025 | Topic 7: Other Private Language Model Adaptations | Room 0.02 |
| 29.01.2025 | Topic 8: Differential Privacy Auditing | Room 0.02 |
| 05.02.2025 | Topic 9: Problems and Open Research Directions in Privacy-Preserving Machine Learning in Foundation Models | Room 0.02 |
Course Staff
- Franziska Boenisch (Instructor)
- Ben Stock (Instructor)