Bias Mitigation in Pipeline

Mitigate bias inside the engineering pipeline rather than at the policy level. Learn pre-processing techniques (re-weighting, re-sampling, suppression), in-processing techniques (constraint-aware training, adversarial debiasing), post-processing techniques (calibration, equalised odds), when each is appropriate (and which kind of fairness it actually buys), and how to verify the mitigation does not introduce other harms (accuracy collapse, group-vs-individual trade-offs, intersectional regressions).