Identifying and Correcting Label Bias in Machine Learning

Abstract

Datasets often contain biases which unfairly disadvantage certain groups, and classifiers trained on such datasets can inherit these biases. In this paper, we provide a mathematical formulation of how this bias can arise. We do so by assuming the existence of underlying, unknown, and unbiased labels which are overwritten by an agent who intends to provide accurate labels but may have biases against certain groups. Despite the fact that we only observe the biased labels, we are able to show that the bias may nevertheless be corrected by re-weighting the data points without changing the labels. We show, with theoretical guarantees, that training on the re-weighted dataset corresponds to training on the unobserved but unbiased labels, thus leading to an unbiased machine learning classifier. Our procedure is fast and robust and can be used with virtually any learning algorithm. We evaluate on a number of standard machine learning fairness datasets and a variety of fairness notions, finding that our method outperforms standard approaches in achieving fair classification.

References in corpus (11)

• Scikit-learn: Machine Learning in Python
• Equality of Opportunity in Supervised Learning
• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness
• Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations
• On Fairness and Calibration
• Satisfying Real-world Goals with Dataset Constraints
• A Reductions Approach to Fair Classification
• Censoring Representations with an Adversary
• Empirical Risk Minimization under Fairness Constraints
• Training Well-Generalizing Classifiers for Fairness Metrics and Other Data-Dependent Constraints

Cited by in corpus (39)

• Fairness in Machine Learning: A Survey
• An Empirical Characterization of Fair Machine Learning For Clinical Risk Prediction
• Does Object Recognition Work for Everyone?
• Towards Intersectionality in Machine Learning: Including More Identities, Handling Underrepresentation, and Performing Evaluation
• Defining the scope of AI regulations
• FairBatch: Batch Selection for Model Fairness
• On Adversarial Bias and the Robustness of Fair Machine Learning
• ET-AL: Entropy-Targeted Active Learning for Bias Mitigation in Materials Data
• Socially Responsible AI Algorithms: Issues, Purposes, and Challenges
• Improving Fairness via Federated Learning
• Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty
• Recovering from Biased Data: Can Fairness Constraints Improve Accuracy?
• FR-Train: A Mutual Information-Based Approach to Fair and Robust Training
• Wasserstein-based fairness interpretability framework for machine learning models
• Counterfactual Reasoning for Fair Clinical Risk Prediction
• Benign Shortcut for Debiasing: Fair Visual Recognition via Intervention with Shortcut Features
• Fair-SSL: Building fair ML Software with less data
• Beyond Low Earth Orbit: Biomonitoring, Artificial Intelligence, and Precision Space Health
• FairIF: Boosting Fairness in Deep Learning via Influence Functions with Validation Set Sensitive Attributes
• Hierarchical Bayesian Models to Mitigate Systematic Disparities in Prediction with Proxy Outcomes
• An Axiomatic Theory of Provably-Fair Welfare-Centric Machine Learning
• Controlling biases and diversity in diverse image-to-image translation
• Physics Enhanced Artificial Intelligence
• FairWASP: Fast and Optimal Fair Wasserstein Pre-processing
• Robust Semantic Interpretability: Revisiting Concept Activation Vectors
• Algorithm Fairness in AI for Medicine and Healthcare
• A Pre-processing Method for Fairness in Ranking
• Metadata Normalization
• The Fair Game: Auditing & Debiasing AI Algorithms Over Time
• Towards the Right Kind of Fairness in AI
• Getting Fairness Right: Towards a Toolbox for Practitioners
• On the Fairness of Machine-Assisted Human Decisions
• Doing good by fighting fraud: Ethical anti-fraud systems for mobile payments
• An Empirical Investigation of Learning from Biased Toxicity Labels
• Unsupervised Recalibration
• Small Business Classification By Name: Addressing Gender and Geographic Origin Biases
• Adaptive Data Debiasing through Bounded Exploration
• Group-based Fair Learning Leads to Counter-intuitive Predictions
• Model-agnostic bias mitigation methods with regressor distribution control for Wasserstein-based fairness metrics