Introduction to Explainable AI
summary
Overview of concrete XAI methods and limits: post-hoc explanations (rule lists, prototypes, counterfactuals, saliency maps) and transparent models versus black‑box inference using LIME and SHAP; legal constraint highlighted is Article 22 GDPR on automated decisions; historical failure example: Amazon’s 2015 hiring model learned gender bias from past CVs; clinical demo shows a convolutional network predicting biological sex from ECG with 89% accuracy and attention maps pointing to the QRS downslope as the influential feature.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:48What is artificial intelligence?
- 2:26What is the problem?
- 3:05What is the objective?
- 4:19What are the challenges?
- 6:11Article 22
- 6:43An example
- 8:06Explainable Models
- 9:13Post-hoc explainability
- 10:42Feature importance toolbox
- 11:28Brief introduction: ECGS
- 12:16Papers: some results
- 13:48Recap
related talks
48mExplainable Artificial Intelligence (XAI): Why, When, and How?
47mML: Explainable Artificial Intelligence explained
35mWhy Do You Need an Explainable Ai?
26mUsing Explainable AI to garner trust and overcome adoption inhibitors
39mThe Search for Transparency and Accountability in the Age of...
01h 29mNavigating ML Explainability: From Theory to Implementation
16mBuilding Trust through Explainable AI
28mThe Need for Explainable AI