About the Course
Deep learning based methods and their uses in various AI applications are very rapidly transforming science and technology. Although there is a tremendous interest in such methods in various fields such as computer vision, pattern recognition, speech recognition, natural language processing, robotics etc., the computational mechanisms for trust and responsibility in AI are still unsolved problems. This course provides an opportunity to students to raise awareness about the social, ethical and policy implications of AI; to study and develop deep learning based methods that are robust, explainable and fair.
This is a graduate-level course taught (CENG7880) at the Dept. of Computer Engineering, METU.
Catalog Description
Ethical concepts and principles of trust and responsibility in AI; computational methods for dependability and robustness in AI; computational methods for explainability in AI; computational methods for bias and fairness in AI.
Background Requirements
- Background in deep learning: The students must have taken CENG403 (Introduction to Deep Learning) or CENG501 (Deep Learning).
- Programming skills in Python.
Instructor
Prof. Dr. Sinan Kalkan
METU Dept. of Computer Engineering and METU Robotics and AI Center
https://user.ceng.metu.edu.tr/~skalkan/
Announcements
- Please fill the following form for submitting your selected papers (D: 16 October): https://forms.gle/A3taWgxoCHumfYfz9
Weekly Coverage
| Week | Topics | Material |
|---|---|---|
| Week 1 (1 Oct) | Introduction to the course and the main concepts in trustworthy and responsible AI | slides |
| Week 2 (8 Oct) | ML/DL Fundamentals | lecture notes |
| Week 3 (15 Oct) | ML/DL Recent Trends | slides |
| Week 4 (22 Oct) | Robust AI: Robustness to Distribution Shifts; Label Shifts & Using Importance Weights | slides and Colab Tutorial on Label Shift (by Ugur Yalcin) |
| Week 5 (29 Oct) | No lectures owing to the Republic Day of Turkey | |
| Week 6 (5 Nov) | Robust AI: Robustness to Covariate Shifts with Importance Weights; Detecting Covariate Shifts; Adversarial Robustness | slides and Colab Tutorial on Covariate Shift (by Ugur Yalcin) |
| Week 7 (12 Nov) | Robust AI: Adversarial Robustness; Adversarial Sample Generation; Adversarial Training; Randomized Smoothing; Certified Robustness; Jailbreaking LLMs | slides |
| Week 8 (19 Nov) | Robust AI: Calibration, Conformal Prediction, Uncertainty Types | slides |
| Week 9 (26 Nov) | Robust AI: Uncertainty Quantification; Explainable AI: Feature Attribution Methods, LIME, SHAP, Gradient-based Saliency Methods | slides |
| Week 10 (3 Dec) | Explainable AI: Quality of feature attribution methods; Counterfactual Explanations | slides |
| Week 10 (3 Dec) | Explainable AI: Guest lecture on Concept Bottleneck Models (CBMs) by Dr Emre Akbas. | slides |
| Week 11 (10 Dec) | Explainable AI: Representation Attribution (Concept Activation Vector); Data Attribution; Explainability in LLMs. | slides |
| Week 12 (17 Dec) | Fairness: Sources of bias; Fairness notions, principles, definitions, criteria, and measures. | slides |
| Week 13 (24 Dec) | Fairness: Fairness algorithms; Fairness in LLMs. | slides |
| Week 14 (31 Dec) | Fairness: Fairness algorithms; Fairness Verification. | slides |