Machine Learning and AI Foundations: Producing Explainable AI (XAI) and Interpretable Machine Learning Solutions

Learn best practices for how to produce explainable AI and interpretable machine learning solutions.

Introduction

• Exploring the world of explainable AI and interpretable machine learning
• Target audience
• What you should know

1. What Are XAI and IML?

• Understanding the what and why your models predict
• Variable importance and reason codes
• Comparing IML and XAI
• Trends in AI making the XAI problem more prominent
• Local and global explanations
• XAI for debugging models
• KNIME support of global and local explanations

2. Why Isolating a Variable’s Contribution Is Difficult

• Challenges of variable attribution with linear regression
• Challenges of variable attribution with neural networks
• Rashomon effect

3. Black Box Model 101

• What qualifies as a black box?
• Why do we have black box models?
• What is the accuracy interpretability tradeoff?
• The argument against XAI

4. Introduction to KNIME for XAI and IML

• Introducing KNIME
• Building models in KNIME
• Understanding looping in KNIME
• Where to find available KNIME support for XAI

5. XAI Techniques: Global Explanations

• Providing global explanations with partial dependence plots
• Using surrogate models for global explanations
• Developing and interpreting a surrogate model with KNIME
• Permutation feature importance
• Global feature importance demo

6. Techniques for Local Explanations

• Developing an intuition for Shapley values
• Introducing SHAP
• Using LIME to provide local explanations for neural networks
• What are counterfactuals?
• KNIME's Local Explanation View node
• XAI View node demonstrating KNIME

7. IML Techniques

• General advice for better IML
• Why feature engineering is critical for IML
• CORELS and recent trends

Conclusion

• Continuing to explore XAI