ReadingList.md

ECS 189G, Spring 2023: a Course on (Social) Fairness of Machine Learning Algorithms

Reading List

Note:

• In some cases, only parts of the listed resource will be required reading.

• Some straightforward parts of the required reading will covered by students on their own, rather than formally in lecture.

• Some of the readings will be very technical, such as the Scutari paper listed below, while some are more qualitative. Both aspects are important.

Prologue

Sets the stage. First example of the fairness issue. Overview of some of the technical terms, but with technical details coming later.

• Pro Publica article

Rates, Proportions, Simpson's Paradox

A firm necessity in the course is a common sense`understanding of proportions, e.g. difference between proportion of x among y vs. proportion of y among x. Amazingly, many people lack this, and in fact, it can indeed be very subtle.

• "natural" solution to Monty Hall problem

  If one carefully names all the events, the problem is actually easy, straightforward.

• UCB admissions discrimination lawsuit

  Could Berkeley, famously liberal, have been discriminating against female applicants for grad school?

• Good for Men, Good for Women, Bad for People

  Title says it all.

Machine Learning Methods, Overfitting, Statistics

• quick tutorial on ML

• quick tutorial on overfitting

• don't use p-values

• technical details of the Pro Publica analysis

• microanalysis of sources of bias in the COMPAS data

General Technical Issues, Measures of Fairness

• article, "Algorithmic Fairness in Computational Medicine

  Good general presentation, not just in medical context. Here we cover the first half.

• paper, "Fairness in Credit Scoring"

  Good general presentation, not just in credit context. First half.

• paper by NM

• COMPAS authors' rebuttal to Pro Publica article

Remedies

• article, "Algorithmic Fairness in Computational Medicine

  Second half.

• paper, "Fairness in Credit Scoring

  Second half.

• Scutari paper

  Highly technical; we will skip the more advanced material, mainly focusing on the intuition.

• NM extension of Scutari paper

ML Fairness in Specific Applications

• NIST on facial recognition

  Short, nontechnical summary.

• data privacy

• Stanford SIEPR analysis of disparity in tax audits