Lectures

You can access the lecture slides here. We will try to upload them before their corresponding classes.

• (xml-1) Introduction
  tl;dr: Attending different notions and properties surrounding Explainability.
  [slides]
  

  Suggested Readings:

  • Ch-3.1 from Interpretable Machine Learning book by Christoph Molnar
  • Towards A Rigorous Science of Interpretable Machine Learning, Doshi-Velez and Kim, 2017
  • Interpretable machine learning: definitions, methods, and applications, Murdoch et al. 2019
  • Explanation in artificial intelligence: Insights from the social sciences, Tim Miller, 2019
  • Examples are not Enough, Learn to Criticize! Criticism for Interpretability, Kim et al. 2016
• (xml-2) Taxonomy, Scope, and Evaluation of Explainability
  tl;dr: Different notions and properties surrounding Explainability.
  [slides]
  

  Suggested Readings:

  • Ch 3.2 to 3.6 from Interpretable Machine Learning book by Christoph Molnar
• (xml-3) Inherently Interpretable Models
  tl;dr: Models with straightforward explanations.
  [slides]
  

  Suggested Readings:

  • Ch 3.1, 4.1 to 4.3 from Introduction to Statistical Learning
  • Ch 5 from Interpretable Machine Learning book by Christoph Molnar
• (xml-4) Local Model-Agnostic Methods - LIME.
  tl;dr: Locally approximating a black-box ML model with an interpretable surrogate
  [slides]
  

  Suggested Readings:

  • LIME, SIGKDD, 2016
  • SHAP, NeurIPS 2017
• (xml-5) Local Model-Agnostic Methods - Counterfactual Explanations.
  tl;dr: Imagining a hypothetical reality to explain predictions of individual instances
  [slides]
  

  Suggested Readings:

  • Counterfactual Explanations without Opening the Black Box (Harvard Journal of Law & Technology, 2018)
  • Multi-Objective Counterfactual Explanations (PPSN, 2020)
  • Counterfactuals and Causability in XAI
• (xml-6) Interpreting Neural Networks (Feature Visualization)
  tl;dr: Methods to understand the units (learned) in DNNs
  [slides]
  

  Suggested Readings:

  • Distil on Visualization
• (xml-7) Interpreting Neural Networks (Pixel Attribution)
  tl;dr: Highlighting the input features (e.g., pixels) relevant for a DNN prediction.
  [slides]
  

  Suggested Readings:

  • Visualising image classification models and saliency maps (ICLRW-2012)
  • Visualizing and understanding CNNs (ECCV-2013)
  • SmoothGrad (2017)
  • Interpreting DNNs is fragile (AAAI-2019)
  • Sanity Checks for Saliency Maps (NeurIPS-2018)
  • The (un) reliability of saliency methods (Explainable AI-2019)
  • Sanity Checks for Saliency Metrics (AAAI-2020)
• (xml-8) Interpreting Neural Networks (Concept-based Explanations)
  tl;dr: Learning a low-dim human-understandable representation that can explain the inference of the DNN.
  [slides]
  

  Suggested Readings:
• (xml-9) Attention and Explanations
  tl;dr: Does the Attention mechanism provide Explanation for the model?
  [slides]
  

  Suggested Readings: