SCADA 2021 Societal Concerns in Algorithms and Data Analysis

Spring Semester 2021

Weizmann Institute of Science

Overview. Machine learning and data analysis have enjoyed tremendous progress in a broad range of domains. These advances hold the promise of great benefits to individuals, organizations, and society as a whole. This progress, however, raises (and is impeded by) a host of concerns. Research has shown that existing machine learning methods can be vulnerable to adversarial attacks, might introduce biases that lead to discrimination and can leak information in a manner that compromises individuals’ privacy. Addressing these vulnerabilities and shortcomings can help society to harness the full power and potential of advances in data science and machine learning.

This course will be devoted to the presentation and discussion of papers that deal with identifying and addressing societal concerns in algorithms, machine learning and data analysis.

NEW: List of student presentations, slides and papers

Before each seminar, students will review background for that lecture. Each student will be expected to present at least one topic in the seminar series.

A website for a past program on these issues can be consulted for further background.

Lecturers: Moni Naor and Guy Rothblum

Time: Wednesdays 14:15-16:00
Zoom Link

Staying in the loop. If you are interested in participating in any or all of these activities, you can join the Google group (this will also give you access to the calendar)

• Lecture 1: group fairness in prediction. Whiteboard notes.

• Lecture 2: individual fairness, multi-calibration. Whiteboard notes.

• Lecture 3: Carpool and the Shapley Value. Whiteboard notes. Lecture Notes on the Carpool problem and Shapley value

  • R. Tijdman, The Chairman Assignment Problem, Discrete Math. 32 (1980)

  • R. Fagin and J. Williams, A fair carpool scheduling algorithm, IBM Journal of Research and Development, 1983.

  • M. Ajtai et al, Fairness in Scheduling, J. Algorithms 29(2), 1998.

  • M. Naor, On Fairness in the Carpool Problem, J. Algorithms 55(1), 2005.

For the Shapley Value see Chapter 12 in Karlin-Peres and any ``standard text" on Game Theory.

Sergiu Hart, SHAPLEY VALUE, The New Palgrave: Dictionary of Economics

• Lecture 4: Voting, Social Choice and Telling the Truth. Whiteboard notes

See Chapter 13 in Karlin-Peres and Chapter 9.2 in Algorithmic Game Theory

• John Geanakoplos, Three Brief Proofs of Arrow's Theorem

• Hervé Moulin, Condorcet's principle implies the no show paradox, 1988

• Felix Brandt, Christian Geist, Dominik Peters, Optimal Bounds for the No-Show Paradox via SAT Solving AAMAS 2016.

• Ehud Friedgut, Gil Kalai, Nathan Keller, and Noam Nisan, A Quantitative Version of the Gibbard-SatterthwaiteTheorem for Three Alternatives. Siam J. Computing 2011

Lecture 5: Causality, Gal Yona. Presentation slides.

• Judea Pearl, Madelyn Glymour and Nicholas P. Jewell, Causal Inference in Statistics: A Primer

• Solon Barocas, Moritz Hardt and Arvind Narayanan, Fairness and Machine Learning Limitation and Opportunities, Chapter 4

• Lily Hu, Disparate Cause: Part I and Part II

• Lily Hu, Direct effects

• Lily Hu & Issa Kohler-Hausmann, What’s Sex got to do with fair machine learning?

Student presentations, papers and slides

Reading Material

• Chapters 10-13 of Game Theory, Alive, by Anna Karlin and Yuval Peres

Resources

• TOC for Fairness, a simons collaboration project

• Simons Institute Cluster on Fairness (2019)

• Conferences:

  • FORC: Foundations of Responsible Computing

  • ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT)

  • ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO) and MD4SG

  • FAT/ML

• Blogs

  • TOC for Fainness Blog

  • Algorithmic Fairness (Suresh)