Overview
Algorithmic fairness involves defining mathematical metrics for 'fairness' and developing techniques to mitigate bias in AI systems.
Common Fairness Metrics
- Demographic Parity: Ensuring the outcome is independent of a protected attribute (e.g., gender).
- Equal Opportunity: Ensuring that 'qualified' individuals from all groups have the same chance of a positive outcome.
Mitigation Strategies
- Pre-processing: Removing bias from the training data.
- In-processing: Adding fairness constraints directly into the model's loss function during training.
- Post-processing: Adjusting the model's predictions after the fact to meet fairness targets.