A Near-Optimal Algorithm for Debiasing Trained Machine Learning Models
arXiv:2106.12887
Abstract
We present a scalable post-processing algorithm for debiasing trained models, including deep neural networks (DNNs), which we prove to be near-optimal by bounding its excess Bayes risk. We empirically validate its advantages on standard benchmark datasets across both classical algorithms as well as modern DNN architectures and demonstrate that it outperforms previous post-processing methods while performing on par with in-processing. In addition, we show that the proposed algorithm is particularly effective for models trained at scale where post-processing is a natural and practical choice.
arXiv comment21 pages, 5 figures
References in corpus (6)
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- Equality of Opportunity in Supervised Learning
- On Calibration of Modern Neural Networks
- Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
- Learning Non-Discriminatory Predictors
- A statistical framework for fair predictive algorithms