Building Trust through Explainable AI
summary
Shows a concrete method for post hoc explanations by using LIME (local interpretable model agnostic explainer) to jiggle features and fit a simple local surrogate so the top weighted features explain a single prediction. Demonstrates code with Google's Inception on an image and highlights superpixels that drive the label (an Alaskan husky example). Applies the approach to a financial advice pipeline for paraprofessionals, argues regulators can audit decisions, and reports the explainer successfully surfaces actionable factors for trust and debugging.
AI-written summary · about
chapters (11)
- 0:00Intro
- 1:20Agenda
- 1:50What we do
- 4:18Tradeoffs
- 5:07Example
- 5:51Lime
- 9:11Model Agnostic
- 10:43Code
- 11:59Trust
- 13:01Line
- 14:06Fix bodies
related talks
30mTowards a more transparent AI - Decrypting AI models using LIME
35mWhy Do You Need an Explainable Ai?
33mExplainable AI in the LIME-light
17mChallenges Towards Production-Ready Explainable Machine Learning
15mIntroduction to Explainable AI
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
39mGuide towards algorithm explainability in machine learning
29mTowards Explainable Machine Learning