Teaching

CS532-T · Topics in AI: Interpretability and Explainability

Term: 2025–26 Winter Term 1

As machine-learning models are increasingly deployed to support decision-makers in high-stakes domains such as biomedicine, healthcare, finance, and law, it becomes critical that those decision-makers understand—or at least trust— the functionality of the models that guide them. This graduate-level course immerses students in the fast-moving field of explainable and interpretable machine learning (XAI).

Over the term we examine foundational ideas in interpretability by discussing seminal papers while probing what interpretability and explainability mean to different stakeholders—from clinicians and biologists seeking actionable explanations to ML engineers debugging model behavior. We traverse the major families of interpretable models and explanation techniques, including prototype-based reasoning, sparse linear surrogates, rule-learning approaches, saliency maps, generalized additive models, and counterfactual analysis.

We further study how interpretability interacts with fairness, robustness, and privacy, and critically evaluate the challenges of demystifying foundation models such as large language models and diffusion models.

The course blends instructor lectures, student paper presentations, and invited talks from leading researchers. A semester-long team project allows students either to pursue new research ideas or to carry out rigorous benchmarking or reproducibility studies.