Data Privacy in AI
Introduce new threat models towards data privacy using AI techniques and how AI techniques can enhance data privacy.
Course Description
This course introduces how the advancement of machine learning and deep learning techniques raise new challenges to protect user data privacy as well as how those new techniques can enhance data privacy.
Topics to be covered
- Introduction (1 class)
- Data Privacy Protection Mechanisms (5 classes)
- Data perturbation (2 classes)
- Cryptography (1 class)
- Anonymization (1 class)
- Discussion (1 class)
- Differential Privacy (6 classes)
- General Differential Privacy (2 classes)
- Metric Differential Privacy (2 classes)
- Context-Aware Indistinguishability (1 class)
- Discussion (1 class)
- AI-based Inference Models (6 classes)
- Classic Inference models (2 classes)
- Machine Learning models (2 classes)
- Deep Neural Networks (1 class)
- Discussion (1 class)
- Generative Adversarial Privacy (4 classes)
- Generative Adversarial Networks (1 class)
- Autoencoders (1 class)
- Generative Adversarial Privacy (1 class)
- Discussion (1 class)
- Projects (6 classes)
- Classic Differential Privacy: Data Privacy Protection in Demographic Analysis (2 classes)
- Metric Differential Privacy: Location Privacy Protection in Spatial Crowdsourcing (2 classes)
- Final Report (2 classes)
Seminar
- “Differential Privacy in AI”
- “Generative Adversarial Privacy”
- “Geo-Indistinguishability in Spatial Crowdsourcing”
Reading Materials
- Deep Learning for Spatio-Temporal Data Mining: A Survey
- Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models
- Deep Learning with Differential Privacy
- Geo-indistinguishability: differential privacy for location-based systems
Projects
- Location Privacy Protection in Spatial Crowdsourcing
- Data Privacy Protection in Demographic Analysis