Learning XAI: Explainable Artificial Intelligence

Introduction

• XAI: the key to fair and transparent AI for all

1. Understanding the AI Landscape

• Bias in AI and its impact
• How does generative AI work
• Challenges of bias mitigation

2. Bias in GenAI

• Bias in data: "Bad data in, bad data out"
• Prompt engineering crash course
• Bias in mathetmical assumptions

3. Bias in Model Training

• Introduction to GenAI model training
• Demographic parity and its applications
• Equal opportunity parity for evaluating fairness
• Equalized odds parity to compare subgroup performance
• HELM
• Red-teaming
• Exercise: Building an evaluation pipeline

4. Counterfactuals in Bias Analysis

• What is a counterfactual?
• Concept of counterfactuals and what-if scenarios
• Applications of counterfactuals for transparency
• Counterfactual modeling for improved fairness
• Exercise: TensorFlow counterfactual

5. Data Sampling for Bias Reduction

• The power of sampling data
• Data validation and evaluation
• Overview of stratified sampling
• Sampling based on demographics
• Random sampling within stratified groups
• Testing model performance across diverse samples
• Exercise: Design sampling criteria with RAI principles

6. Post-Training Model Constraints

• Introduction to model constraints to control bias
• Power of flagging, pruning, and nudging
• Using human-in-the-loop tools
• Exercise: Model constraint pipeline

7. Case Study: Breakdown of a GenAI Story

• Explore an example of GenAI output

Conclusion

• Continuing your XAI learning journey