Awesome-Fairness-in-AI

A curated, but probably biased and incomplete, list of awesome Fairness in AI resources.

If you want to contribute to this list, feel free to pull a request. Also you can contact Mengnan Du from the Data Lab at Texas A&M University through email: dumengnan@tamu.edu, or Twitter @DuMNCH.

What is Fairness in AI?

AI algorithms are increasingly being used in high-stake decision making applications that affect individual lives. However, AI might exhibit algorithmic discrimination behaviors with respect to protected groups, potentially posing negative impacts on individuals and society.

Fairness in AI (FAI) aims to build fair and unbiased AI/machine learning systems, that ensure the benefits are broadly available across all segments of society. Specific topics include but are not limited to: theoretical understanding of algorithmic bias, defining measurements of fairness, detection of adverse biases, and developing mitigation strategies.

Table of Contents

• Review and General Papers
• Measurements of Fairness
• Demonstration of Bias Phemomenon in Various Applications
  • Bias in Machine Learning Models
  • Bias in Representations
• Mitigation of Unfairness
  • Mitigation of Machine Learning Models
    • Adversarial Learning
    • Calibration
    • Incorporating Priors into Feature Attribution
    • Data Collection
    • Other Mitigation Methods
  • Mitigation of Representations
• Fairness Packages and Frameworks
• Conferences
• Other Fairness Relevant Interpretability Resources

Review and General Papers

• Fairness in Deep Learning: A Computational Perspective
• The Measure and Mismeasure of Fairness: A Critical Review of Fair Machine Learning
• Fairness and machine learning
• A Survey on Bias and Fairness in Machine Learning
• The Frontiers of Fairness in Machine Learning
• Ensuring fairness in machine learning to advance health equity
• Mitigating Gender Bias in Natural Language Processing: Literature Review
• Fairness in Recommender Systems
• Implementations in Machine Ethics: A Survey

Measurements of Fairness

• On Formalizing Fairness in Prediction with Machine Learning
• Equality of Opportunity in Supervised Learning
• Certifying and removing disparate impact
• Does mitigating ML's impact disparity require treatment disparity?
• Putting Fairness Principles into Practice: Challenges, Metrics, and Improvements
• Beyond Parity: Fairness Objectives for Collaborative Filtering
• 50 Years of Test (Un)fairness: Lessons for Machine Learning
• Fairness Definitions Explained
• Algorithmic Fairness
• Bias in data‐driven artificial intelligence systems—An introductory survey
• Fairness is not Static: Deeper Understanding of Long Term Fairness via Simulation Studies
• Delayed Impact of Fair Machine Learning

Demonstration of Bias Phemomenon in Various Applications

Bias in Machine Learning Models

• Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification
• Deep Learning for Face Recognition: Pride or Prejudiced?
• Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems
• Demographic Dialectal Variation in Social Media: A Case Study of African-American English
• Feature-Wise Bias Amplification
• ConvNets and ImageNet Beyond Accuracy: Understanding Mistakes and Uncovering Biases

Bias in Representations

• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• Gender Bias in Contextualized Word Embeddings
• Assessing Social and Intersectional Biases in Contextualized Word Representations

Mitigation of Unfairness

Mitigation of Machine Learning Models

• Adversarial Learning

  • Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations
  • Achieving Fairness through Adversarial Learning: an Application to Recidivism Prediction
  • Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations
  • Mitigating Unwanted Biases with Adversarial Learning
  • Adversarial Removal of Demographic Attributes from Text Data
  • Compositional Fairness Constraints for Graph Embeddings

• Calibration

  • Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints
  • Equality of Opportunity in Supervised Learning

• Incorporating Priors into Feature Attribution

  • Incorporating priors with feature attribution on text classification
  • Women Also Snowboard: Overcoming Bias in Captioning Models
  • Learning Credible Deep Neural Networks with Rationale Regularization

• Data Collection

  • Why Is My Classifier Discriminatory?
  • Incorporating Dialectal Variability for Socially Equitable Language Identification
  • REPAIR: Removing Representation Bias by Dataset Resampling

• Other Mitigation Methods

  • Mitigating Bias in Gender, Age and Ethnicity Classification: a Multi-Task Convolution Neural Network Approach
  • InclusiveFaceNet: Improving Face Attribute Detection with Race and Gender Diversity
  • Reducing Gender Bias in Abusive Language Detection
  • Fairness-aware Learning through Regularization Approach
  • Fairness Constraints: Mechanisms for Fair Classification
  • Penalizing Unfairness in Binary Classification
  • Fairness Constraints:A Flexible Approach for Fair Classification
  • A General Framework for Fair Regression
  • Fair Regression: Quantitative Definitions and Reduction-based Algorithms

Mitigation of Representations

• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
• Learning Gender-Neutral Word Embeddings
• Flexibly Fair Representation Learning by Disentanglement
• Turning a Blind Eye: Explicit Removal of Biases and Variation from Deep Neural Network Embeddings

Fairness Packages and Frameworks

• AI Fairness 360
• fairlearn: Fairness in machine learning mitigation algorithms
• algofairness
  • fairness: Benchmarking of fairness aware machine learning algorithms
• FairSight: Visual Analytics for Fairness in Decision Making
• GD-IQ: Spellcheck for Bias (code not available)
• aequitas: Bias and Fairness Audit Toolkit
• CERTIFAI: A Common Framework to Provide Explanations and Analyse the Fairness and Robustness of Black-box Models
• ML-fairness-gym: Google's implementation based on OpenAI's Gym
• fairness-indicators: Tensorflow's Fairness Evaluation and Visualization Toolkit
• scikit-fairness

Conferences

• FAT*: ACM Conference on Fairness, Accountability, and Transparency
• FATML: Fairness, Accountability, and Transparency in Machine Learning Workshop
• AIES: AAAI/ACM Conference on AI, Ethics, and Society

Other Fairness Relevant Interpretability Resources

• Awesome machine learning interpretability resources
• The Mythos of Model Interpretability
• Towards A Rigorous Science of Interpretable Machine Learning
• Techniques for Interpretable Machine Learning
• Methods for Interpreting and Understanding Deep Neural Networks