Course Details

Subject {L-T-P / C} : CS6608 : Explainable AI { 3-0-0 / 3}

Subject Nature : Theory

Coordinator : Puneet Kumar Jain

Syllabus

UNIT I: Introduction to Explainable AI
Science of Interpretable Machine Learning, Motivation, Challenges, and Mythos of Model Interpretability, Human Factors in Explainability, Interpreting Interpretability

UNIT II: Post hoc Explanations
Explaining the Predictions of Any Classifier, Pitfalls, Challenges, and Evaluation of Feature Attributions, LIME and SHAP, OpenXAI, The Disagreement Problem in Explainable Machine Learning, Counterfactual Explanations (or) Algorithmic Recourse, Learning Model-Agnostic Counterfactual Explanations for Tabular Data

UNIT III: Attention and Concept Based Explanations
Quantifying Interpretability of Deep Visual Representations, Interpretability Beyond Feature Attribution, Data Attribution and Interactive Explanation, Equitable Valuation of Data, Explainable Active Learning (XAL), Theory of Explainability and Interpreting Generative Models

UNIT IV: Explainability for Fair Machine Learning
Connections with Robustness, Privacy, Fairness, and Unlearning, Right to Explanation and the Right to be Forgotten, Fairness via Explanation Quality, Mechanistic Interpretability and Compiled Transformers, Understanding and Reasoning in Large Language Models

Course Objectives

Course Outcomes

The course aims to achieve the following outcomes:
CO1: Students will effectively apply interpretability techniques, such as feature importance analysis, rule-based models, and model-agnostic methods, to enhance the transparency of machine learning models.
CO2: Students will possess the ability to evaluate and compare different interpretability methods.
CO3: Students will develop a critical understanding of the trade-offs between model complexity and interpretability in diverse scenarios.
CO4: Demonstrating practical skills, students will apply explainability techniques to real-world datasets and challenges.
CO5: Students will gain awareness of the ethical implications related to AI transparency and interpretability.

Essential Reading

Supplementary Reading

Journal and Conferences