Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations

Abstract

How can we learn a classifier that is "fair" for a protected or sensitive group, when we do not know if the input to the classifier belongs to the protected group? How can we train such a classifier when data on the protected group is difficult to attain? In many settings, finding out the sensitive input attribute can be prohibitively expensive even during model training, and sometimes impossible during model serving. For example, in recommender systems, if we want to predict if a user will click on a given recommendation, we often do not know many attributes of the user, e.g., race or age, and many attributes of the content are hard to determine, e.g., the language or topic. Thus, it is not feasible to use a different classifier calibrated based on knowledge of the sensitive attribute. Here, we use an adversarial training procedure to remove information about the sensitive attribute from the latent representation learned by a neural network. In particular, we study how the choice of data for the adversarial training effects the resulting fairness properties. We find two interesting results: a small amount of data is needed to train these adversarial models, and the data distribution empirically drives the adversary's notion of fairness.

Presented as a poster at the 2017 Workshop on Fairness, Accountability, and Transparency in Machine Learning (FAT/ML 2017)

References in corpus (1)

• Censoring Representations with an Adversary

Cited by in corpus (101)

• Fairness in Machine Learning: A Survey
• Fairness in Machine Learning
• 50 Years of Test (Un)fairness: Lessons for Machine Learning
• A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications
• The Frontiers of Fairness in Machine Learning
• Learning Adversarially Fair and Transferable Representations
• Personalized Counterfactual Fairness in Recommendation
• Connecting User and Item Perspectives in Popularity Debiasing for Collaborative Recommendation
• EDITS: Modeling and Mitigating Data Bias for Graph Neural Networks
• Identifying and Correcting Label Bias in Machine Learning
• Achieving Fairness through Adversarial Learning: an Application to Recidivism Prediction
• Measuring and Reducing Gendered Correlations in Pre-trained Models
• Fighting Fire with Fire: Using Antidote Data to Improve Polarization and Fairness of Recommender Systems
• Stereotypical Bias Removal for Hate Speech Detection Task using Knowledge-based Generalizations
• A Comprehensive Survey on Trustworthy Graph Neural Networks: Privacy, Robustness, Fairness, and Explainability
• Empirical Risk Minimization under Fairness Constraints
• Data synthesis and adversarial networks: A review and meta-analysis in cancer imaging
• InclusiveFaceNet: Improving Face Attribute Detection with Race and Gender Diversity
• Differential Privacy Has Disparate Impact on Model Accuracy
• Using generative AI to investigate medical imagery models and datasets
• Towards Equity and Algorithmic Fairness in Student Grade Prediction
• Fairness without Demographics through Adversarially Reweighted Learning
• Mitigating Gender Bias in Natural Language Processing: Literature Review
• FairGAN: Fairness-aware Generative Adversarial Networks
• Towards Fair Classifiers Without Sensitive Attributes: Exploring Biases in Related Features
• Probably Approximately Metric-Fair Learning
• Provably Fair Representations
• Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations
• Hurtful Words: Quantifying Biases in Clinical Contextual Word Embeddings
• Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty
• Inherent Tradeoffs in Learning Fair Representations
• Transfer of Machine Learning Fairness across Domains
• Putting Fairness Principles into Practice: Challenges, Metrics, and Improvements
• Algorithmic Decision Making with Conditional Fairness
• Robustness Disparities in Commercial Face Detection
• DVR: Micro-Video Recommendation Optimizing Watch-Time-Gain under Duration Bias
• Whither Bias Goes, I Will Go: An Integrative, Systematic Review of Algorithmic Bias Mitigation
• Reducing Sentiment Bias in Language Models via Counterfactual Evaluation
• Wasserstein Fair Classification
• Review of Mathematical frameworks for Fairness in Machine Learning
• Towards Fair Knowledge Transfer for Imbalanced Domain Adaptation
• Fair Representation: Guaranteeing Approximate Multiple Group Fairness for Unknown Tasks
• Estimating and Improving Fairness with Adversarial Learning
• Conditional Learning of Fair Representations
• Adversarial training approach for local data debiasing
• Fair Generative Modeling via Weak Supervision
• Adversarial representation learning for private speech generation
• Towards Accuracy-Fairness Paradox: Adversarial Example-based Data Augmentation for Visual Debiasing
• Say No to the Discrimination: Learning Fair Graph Neural Networks with Limited Sensitive Attribute Information
• Removing biased data to improve fairness and accuracy
• Contrastive Examples for Addressing the Tyranny of the Majority
• Fairness-Aware Node Representation Learning
• Adversarial Removal of Demographic Attributes from Text Data
• Latent Adversarial Debiasing: Mitigating Collider Bias in Deep Neural Networks
• AI Fairness via Domain Adaptation
• Algorithmic Bias and Data Bias: Understanding the Relation between Distributionally Robust Optimization and Data Curation
• Removing Disparate Impact of Differentially Private Stochastic Gradient Descent on Model Accuracy
• Learning Fair and Transferable Representations
• Discovering Fair Representations in the Data Domain
• The Fairness-Accuracy Pareto Front
• Application-driven Privacy-preserving Data Publishing with Correlated Attributes
• Comparing Human and Machine Bias in Face Recognition
• Adversarial Training for Multilingual Acoustic Modeling
• Toward a better trade-off between performance and fairness with kernel-based distribution matching
• An Effective, Robust and Fairness-aware Hate Speech Detection Framework
• Technical Challenges for Training Fair Neural Networks
• Impossibility results for fair representations
• Using Positive Matching Contrastive Loss with Facial Action Units to mitigate bias in Facial Expression Recognition
• Understanding and Mitigating Accuracy Disparity in Regression
• Demographics Should Not Be the Reason of Toxicity: Mitigating Discrimination in Text Classifications with Instance Weighting
• Avoiding Resentment Via Monotonic Fairness
• Interpretable bias mitigation for textual data: Reducing gender bias in patient notes while maintaining classification performance
• Distributed generation of privacy preserving data with user customization
• Technologies for Trustworthy Machine Learning: A Survey in a Socio-Technical Context
• Deletion-Robust Submodular Maximization at Scale
• Automatic Fairness Testing of Neural Classifiers through Adversarial Sampling
• Removing Spurious Features can Hurt Accuracy and Affect Groups Disproportionately
• Adversarial Recommendation: Attack of the Learned Fake Users
• Adversarial Stacked Auto-Encoders for Fair Representation Learning
• Fairness in Deep Learning: A Computational Perspective
• Additive Adversarial Learning for Unbiased Authentication
• Towards a Measure of Individual Fairness for Deep Learning
• Fairness-aware Personalized Ranking Recommendation via Adversarial Learning
• RAWLSNET: Altering Bayesian Networks to Encode Rawlsian Fair Equality of Opportunity
• REDAT: Accent-Invariant Representation for End-to-End ASR by Domain Adversarial Training with Relabeling
• Generating private data with user customization
• Equality of opportunity in travel behavior prediction with deep neural networks and discrete choice models
• Debiasing Convolutional Neural Networks via Meta Orthogonalization
• Adaptation and Generalization for Unknown Sensitive Factors of Variations
• Towards Fair Deep Anomaly Detection
• Learning Fair Representations for Recommendation: A Graph-based Perspective
• Towards Model-Agnostic Post-Hoc Adjustment for Balancing Ranking Fairness and Algorithm Utility
• Fairness-Aware Online Personalization
• DADI: Dynamic Discovery of Fair Information with Adversarial Reinforcement Learning
• BriarPatches: Pixel-Space Interventions for Inducing Demographic Parity
• Enhancing Model Robustness and Fairness with Causality: A Regularization Approach
• Towards Auditability for Fairness in Deep Learning
• Topics to Avoid: Demoting Latent Confounds in Text Classification
• Quantifying point cloud realism through adversarially learned latent representations
• HyperFair: A Soft Approach to Integrating Fairness Criteria
• On the Fairness of 'Fake' Data in Legal AI