CSC 2541F: AI and Ethics: Mathematical Foundations and Algorithms
Fall, 2019

GRADES AND MARKING:

Class attendance/participation (10% of grade)
2 assignments (each worth 25% of grade)
Poster presentation and write-up (40% of grade)

HOMEWORK ASSIGNMENTS:

• Homework 1 (Due Oct 8, 3pm)
• Homework 2 (Due Nov 4, 3pm)

LECTURE NOTES AND READINGS:

Lecture 1 (Sept 10): Societal Impacts of AI; Introduction to Fairness

Lecture 2 (Sept 17): Definitions and Tradeoffs

• PRIMARY: Fairness and Machine Learning, Chapter 2, Barocas, S., Hardt, M. and Narayanan, A.
• Inherent Tradeoffs in the Determination of Risk Scores, Kleinberg, J., Mullainathan, S., and Raghavan, M.
• Fair prediction with disparate impact, Chouldechova, A.
• Inherent Tradeoffs in Learning Fair Representation, Zhao, H., and Gordon, G.
• On the Impossibility of Fairness, Friedler, S., Scheidegger, C., and Venkatasubramanian, S.
• Fairness Under Composition, Dwork, C. and Ilvento, C.
• Simplicity Creates Inequity: Implications for Fairness, Stereotypes and Interpretability, Kleinberg, J. and Mullainathan, S.
• Fairness-Efficiency Tradeoffs in Dynamic Fair Division, Zeng, D. and Psomas, A.

Lecture 3 (Sept 24): Sources of Bias
• PRIMARY: Semantics derived automatically from language corpora contain human-like biases, Caliskan, A., Bryson, J. and Narayanan, A..
• PRIMARY: A large-scale analysis of racial disparities in police stops across the United States, Pierson, E., et al.
• Unbiased look at dataset bias, Torralba, A. and Efros, A.
• Why is my classifier discriminatory?, Chen, I., Johansson, F. and Sontag, D.
• Man is to computer programmer as woman is to homemaker? Debiasing word embeddings, Bolukbasi, et al.
• Understanding undesirable word embedding associations, Ethayarajh, K., Duvenaud, D., and Hirst, G.

Lecture 4 (Oct 1): Fairness Mechanisms
• PRIMARY: Fairness through awareness, Dwork, C. et al.
• PRIMARY: Learning adversarially fair and transferable representations, Madras, D. et al.
• Optimized data pre-processing for discrimination prevention, Calmon, F. et al.
• Certifying and removing disparate impact, Feldman, M. et al.
• Invariant representations without adversarial training, Moyer, D. et al.
• Learning controllable fair representations, Song, J. et al.

Lecture 5 (Oct 8): Introduction to Differential Privacy
• PRIMARY: The Algorithmic Foundations of Differential Privacy, Chapters 1-3, Dwork, C., and Roth, A.


Lecture 6 (Oct 15): Making ML Systems Private (Part 1; Part 2)
• PRIMARY: What Can We Learn Privately?, Kasiviswanathan, S. et al.
• PRIMARY: Tutorial on Differential Privacy, Chaudhuri, K. and Sarwate, A.
• PRIMARY: Deep learning with differential privacy, Abadi, M., et al.
• Scalable private learning with PATE, Papernot, N., et al.
• Privacy amplification by subsampling, Smith, A.
• Semi-supervised knowledge transfer for deep learning from private training data, Papernot, N., et al.
• Learning in a large function space: Privacy-preserving mechanisms for {SVM} learning, Benjamin, I. et al.
• Differentially private empirical risk minimization, Chaudhuri, K. et al.
• The secret sharer: Measuring unintended neural network memorization extracting secrets, Carlini, N. et al.
• Model inversion attacks that exploit confidence information and basic countermeasures, Fredrikson, M. et al.

Lecture 7 (Oct 22): Causality and Fairness
• Avoiding discrimination through causal reasoning Kilbertus, N., et al
• Counterfactual fairness, Kusner, M., et al
• To predict and serve?, Lum, K. and Isaac, W.
• Residual unfairness in fair machine learning from prejudiced data, Zhou, A. and Kallus, N.

Lecture 8 (Oct 29): Fairness in Dynamic Settings
• Woulda, coulda, shoulda: Counterfactually-guided policy search, Buesing, I. et al.
• SIREN: A Simulation Framework for Understanding the Effects of Recommender Systems in Online News Environments, Bountouridis, R. et al.
• Counterfactual off-policy evaluation with Gumbel-max structural causal models, Oberst, M. and Sontag, D
• Causal modeling for fairness in dynamical systems, Creager, E. et al
• Fairness without demographics in repeated loss minimization, Hashimoto, T. et al.
• A short-term intervention for long-term fairness in the labor market, Hu, L. and Chen, Y.
• Fairness in reinforcement learning Jabbari, S., et al
• Downstream effects of affirmative action Kannan, S., et al

Lecture 9 (Nov 12): Extending the Formulation: Multi-group Fairness
• Multicalibration: Calibration for the (Computationally-Identifiable) Masses, Herbert-Johnson, M., et al
• Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness, Kearns, M., et al
• Multiaccuracy: Black-box post-processing for fairness in classification, Kim, M., et al

Lecture 10 (Nov 19): Fairness in Economics & Game Theory
• PRIMARY: Designing fairly fair classifiers via economic fairness notions , Hossain, S., et al.
• PRIMARY: Envy-free classification , Balcan, M-F., et al.
• Protecting the protected group: Circumventing harmful fairness, Ben-Porat, O., et al.
• Fairness without harm: Decoupled classifiers with preference guarantees, Ustun, B., et al.
• From parity to preference-based notions of fairness in classification, Zafar, M., et al.

Lecture 11 (Nov 26): Beyond Classification: Other Domains & Viewpoints

• Dissecting racial bias in an algorithm used to manage the health of populations, Powers, B., et al
• Racial Disparities and Mistrust in End-of-Life Care, Boag, W., et al
• Can AI Help Reduce Disparities in General Medical and Mental Health Care?, Chen, I., et al

Other Topics: Learning the Metric

• Eliciting and Enforcing Subjective Individual Fairness, Jung, M., et al
• Metric learning for individual fairness, Ilvento, C.
• Probably approximately metric fair learning}, Rothblum, G. and Gal, Y.