Why and How to Build Explainability into your ML Workflow
summary
Companies report nearly half lack trust in AI and regulators invoke Article 22 (GDPR), so explainability becomes a compliance and risk tool. Practical methods shown include SHAP and integrated gradients plus ablation and gradient techniques, with a striking medical example where a high‑AUC model learns radiologist markup as a spurious signal. Concrete outputs: feature attributions (e.g. FICO contributes ~30% to a loan score), dashboards and APIs from a Fiddler Labs explainability engine that plugs into training and production.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 5:00How to build explainability
- 7:30What is attribution
- 10:35What is the baseline
- 12:50What about forestbased models
- 14:55Credit request example
- 18:05How to evaluate explainability
- 18:40Why do we need explanations
- 19:15Causality
- 19:55Transferability
- 20:35Providing more information
- 21:05Summary
- 21:20Explainability Platform
- 21:50Why am I seeing this
- 22:35Our vision
related talks
01h 29mNavigating ML Explainability: From Theory to Implementation
35mWhy Do You Need an Explainable Ai?
41mThe Explainability Problem: Towards Understanding AI
32mOn the Importance of Explainability
39mGuide towards algorithm explainability in machine learning
39mExplainable AI as a Conventional Data Analysis Tool
15mIntroduction to Explainable AI
28mThe Need for Explainable AI