19333101
Lecture
SoSe 23: Cybersecurity and AI II
Gerhard Wunder
Comments
The lecture is part II of the cycle Cybersecurity I-IV (data protection, explainability, robustness/attacks, certification) and includes:
- Introduction, Motivation, Definitions
Taxonomy and assumptions, brief overview of ideas from different categories, representation of explanation results, saliency maps, feature importance, first applications - Black-box (model-agnostic) explanations
Additive feature attribution method and properties, LIME and other variants, SHAP, Shapley values, from local explanation to global understanding Implementation details regarding neighborhood construction (on-manifold explanations), risks of random perturbations - White-box (model-specific) explanations
LRP, DTD, DeepLIFT, Grad-CAM, Counterfactual explanations - Information-theoretic explanation methods
Information decomposition, causality, theory of representation learning - Application and Implementation
Debugging, model extraction, challenges, trade-off (e.g., explainability vs. privacy)
14 Class schedule
Regular appointments
Tue, 2023-04-18 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-04-25 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-05-02 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-05-09 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-05-16 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-05-23 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-05-30 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-06-06 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-06-13 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-06-20 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-06-27 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-07-04 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-07-11 12:00 - 14:00
Cybersecurity and AI II
Tue, 2023-07-18 12:00 - 14:00
Cybersecurity and AI II