Learning Controllable Fair Representations

Abstract

Learning data representations that are transferable and are fair with respect to certain protected attributes is crucial to reducing unfair decisions while preserving the utility of the data. We propose an information-theoretically motivated objective for learning maximally expressive representations subject to fairness constraints. We demonstrate that a range of existing approaches optimize approximations to the Lagrangian dual of our objective. In contrast to these existing approaches, our objective allows the user to control the fairness of the representations by specifying limits on unfairness. Exploiting duality, we introduce a method that optimizes the model parameters as well as the expressiveness-fairness trade-off. Empirical evidence suggests that our proposed method can balance the trade-off between multiple notions of fairness and achieves higher expressiveness at a lower computational cost.

arXiv commentAISTATS 2019, fixed a typo

Cited by in corpus (32)

• An Empirical Characterization of Fair Machine Learning For Clinical Risk Prediction
• Invariant Representations without Adversarial Training
• Overlearning Reveals Sensitive Attributes
• TIPRDC: Task-Independent Privacy-Respecting Data Crowdsourcing Framework for Deep Learning with Anonymized Intermediate Representations
• FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders
• A Theory of Usable Information Under Computational Constraints
• Inherent Tradeoffs in Learning Fair Representations
• Learning Certified Individually Fair Representations
• Improving Zero-shot Voice Style Transfer via Disentangled Representation Learning
• FACT: A Diagnostic for Group Fairness Trade-offs
• Review of Mathematical frameworks for Fairness in Machine Learning
• Fair Generative Modeling via Weak Supervision
• Rényi Fair Inference
• Layer-wise Characterization of Latent Information Leakage in Federated Learning
• Newton-type Methods for Minimax Optimization
• Learning Invariant Representations and Risks for Semi-supervised Domain Adaptation
• Guided Variational Autoencoder for Disentanglement Learning
• Tight Mutual Information Estimation With Contrastive Fenchel-Legendre Optimization
• AI Fairness via Domain Adaptation
• Multi-label Contrastive Predictive Coding
• On the Fairness of Disentangled Representations
• Impossibility results for fair representations
• Distributed generation of privacy preserving data with user customization
• Robust Fairness-aware Learning Under Sample Selection Bias
• Privacy-Constrained Policies via Mutual Information Regularized Policy Gradients
• Fair Normalizing Flows
• Generating private data with user customization
• Fairness Constraints in Semi-supervised Learning
• D2C: Diffusion-Denoising Models for Few-shot Conditional Generation
• Adaptation and Generalization for Unknown Sensitive Factors of Variations
• Explaining Algorithmic Fairness Through Fairness-Aware Causal Path Decomposition
• Improving Compositionality of Neural Networks by Decoding Representations to Inputs