The Three Stages of Bias Mitigation
Learn when and how to intervene at each stage of the ML pipeline — pre-processing, in-processing, and post-processing — with practical strategies ranked by effectiveness and difficulty.
Three-Stage Framework
Explore the techniques available at each stage. Click a stage tab to see its strategies, then expand any technique for detailed implementation guidance.
Mitigation Strategy Picker
Select your HR scenario to receive recommended mitigation strategies ranked by effectiveness and implementation difficulty.
Stage Comparison Tool
Compare the three mitigation stages side by side across key dimensions to understand when to use which approach.
Key insight: No single mitigation technique is sufficient. The most robust approach layers interventions across all three stages — cleaning the data before training, constraining the model during training, and adjusting outputs after training. Think of it as defence in depth: each layer catches biases that the others miss.