Course Information
The aim of this course is to provide a systematic and research-oriented overview of Explainable Artificial Intelligence (XAI), with a strong focus on interpretability methods for complex models, evaluation of explanations, and practical use of XAI in high-stakes domains.
The course emphasizes:
1. Conceptual foundations of interpretability
2. Model-agnostic and model-specific explanation techniques
3. Faithfulness, robustness, and evaluation of explanations
4. XAI for deep learning systems
5. The role of XAI in trustworthy, accountable, and human-centered AI
Learning outcomes
After completing the course the student
1) Explain and contrast major paradigms of interpretability in machine learning (intrinsic vs post-hoc, global vs local).
2) Apply modern XAI techniques (e.g., feature attribution, concept-based, counterfactual, and surrogate models) to complex ML and DL systems.
3) Critically analyze XAI research with respect to faithfulness, usability, and limitations.
4) Evaluate explanations using quantitative, qualitative, and human-centered criteria.
Brief description of content
This seminar-based course explores both foundational and state-of-the-art methods in Explainable AI. Students will study, present, and debate influential and recent research papers in XAI. Practical understanding is reinforced through hands-on experiments with explainability methods applied to real models and datasets.
The course focuses on:
1. Explanation-driven analysis,
2. Comparative evaluation of XAI methods, and
3. Mini research-style assignments aligned with current XAI research practice.
Passing Criteria
In order to pass the course, the student must have to provide:
1. Presentation of two XAI research papers (individual or paired)
2. One practical XAI project
3. Independent Assignments
Assessment methods
FInal assessment is non-differentiated (pass, fAil, not present).
Presentation:
1. Quality and depth of paper analysis
2. Ability to critically discuss strengths and limitations of XAI methods
Project:
Students must complete a small project involving the application of at least two different XAI methods to a trained machine learning or deep learning model. The project should include an explanation-driven analysis, a comparison of XAI methods, and a short written report describing the methodology, observations, and conclusions.
Grading Requirements
Collect at least 70 points. The points come from active participation in seminars, assigned homeworks and projects.
Deadlines
All deadlines in the course are strict deadlines.
Course Forum
Moodle Link
Forum and discussions will be held in Moodle.
Contact
Lecturer:
Radwa Mohamed El Emam El Shawi (radwa.elshawi@ut.ee)
Teaching Assistants:
Mehak Mushtaq Malik (mehak1@ut.ee)
Abdesselam Ferdi (abdesselam.ferdi@ut.ee)