• Corpus ID: 251719250

Algorithm Design: Fairness and Accuracy

@inproceedings{Liang2021AlgorithmDF,
  title={Algorithm Design: Fairness and Accuracy},
  author={Annie Liang and Jay Lu and Xiaosheng Mu},
  year={2021}
}
Algorithms are widely used to guide high-stakes decisions, from medical recommendations to loan approvals. Designers are increasingly optimizing not only for accuracy but also “fairness” i.e. how much accuracy varies across different subgroups. We define and characterize a fairness-accuracy frontier, consisting of the optimal points across a broad range of criteria for trading off fairness and accuracy. Our results identify how the algorithm’s inputs govern the shape of this frontier, showing… 

References

SHOWING 1-10 OF 44 REFERENCES

Cardinal Utility in Welfare Economics and in the Theory of Risk-taking

From most branches of economics the concept of cardinal utility has been eliminated as redundant since ordinal utility has been found to suffice for doing the job. Cardinal utility has been kept only

Equality or Priority

This is the text of The Lindley Lecture for 1991, given by Derek Parfit (1942-2013), a British philosopher.

A Theory of Justice

  • J. Rawls
  • Law
    Princeton Readings in Political Thought
  • 1971
John Rawls is Professor Emeritus at Harvard University. He is the author of the well-known and path breaking A Theory of Justice (Harvard, 1971) and the more recent work Political Liberalism

Equality or Priority?" in The Ideal of Equality

  • 2002

Dissecting racial bias in an algorithm used to manage the health of populations

TLDR
It is suggested that the choice of convenient, seemingly effective proxies for ground truth can be an important source of algorithmic bias in many contexts.

Racial Discrimination.

  • A. Klass
  • Law
    Canadian Medical Association journal
  • 1963
TLDR
The Commission held that no further action was called for, since the applicants had subsequently been admitted to the United Kingdom, and observed that discrimination based on race could in certain circumstances amount to degrading treatment.

Equality of Opportunity in Supervised Learning

TLDR
This work proposes a criterion for discrimination against a specified sensitive attribute in supervised learning, where the goal is to predict some target based on available features and shows how to optimally adjust any learned predictor so as to remove discrimination according to this definition.

The Ethical Algorithm: The Science of Socially Aware Algorithm Design

  • 2019

Fair prediction with disparate impact: A study of bias in recidivism prediction instruments

TLDR
It is demonstrated that the criteria cannot all be simultaneously satisfied when recidivism prevalence differs across groups, and how disparate impact can arise when an RPI fails to satisfy the criterion of error rate balance.

Inherent Trade-Offs in the Fair Determination of Risk Scores

TLDR
Some of the ways in which key notions of fairness are incompatible with each other are suggested, and hence a framework for thinking about the trade-offs between them is provided.