Explaining a Machine Learning blackbox
summary
Covers methods for interpreting opaque machine learning models by demonstrating model-agnostic and model-specific explanation techniques such as LIME, SHAP, and feature importance approaches, applied to real-world prediction tasks. Shows practical workflows and tools for generating local and global explanations to help stakeholders understand and trust black-box models.
AI-written summary · about
chapters (24)
- 0:00Intro
- 1:40What is interpretability?
- 2:47Interpretability is essential for
- 3:35Accuracy vs. Interpretability
- 4:12Text Classification
- 4:59Loan Approval
- 5:22Husky or Malamute?
- 7:40Why it's important? Bias and Ethics. Two Drug Possession Arrests
- 10:02Why it's important? Regulations.
- 11:54Dimensions of Interpretability
- 14:01Interpretability before model creation
- 14:58Interpretability during model creation
- 15:45Linear models
- 16:30Decision trees
- 17:33Interpretability after model creation
- 18:24LIME. Two-class case
- 19:59LIME. Multi-class case
- 21:15LIME. Tabular data
- 21:50LIME. Image classification
- 24:44Explaining a prediction with LIME
- 27:53Shap (SHapley Additive explanations) SHAP connects game theory with local explanations, and representing the only possible consistent and locally accurate additive feature attribution method based on expectations.
- 28:22Sentiment Analysis with Shap
- 32:46Microsoft Azure Machine Learning Service (InterpretML)
- 33:29References
related talks
28mExplaining Machine Learning Models with LLMs
29mTowards Explainable Machine Learning
01h 36mOpen the Black Box: an Introduction to Model Interpretability with LIME and SHAP
45mExplainable Machine Learning Explained
30mKeep Calm and Trust your Model - On Explainability of Machine Learning Models
16mBuilding Trust through Explainable AI
17mChallenges Towards Production-Ready Explainable Machine Learning
44mInterpretable Machine Learning with Python Examples