Training Well-Generalizing Classifiers for Fairness Metrics and Other Data-Dependent Constraints

By Andrew Cotter, Maya R. Gupta, Heinrich Jiang, Nati Srebro, Karthik Sridharan, Serena Wang, Blake E. Woodworth, Seungil You. 2018

119 citations recorded by the source corpus.

Classifiers can be trained with data-dependent constraints to satisfy fairness goals, reduce churn, achieve a targeted false positive rate, or other policy goals. We study the generalization performance for such constrained optimization problems, in terms of how well the constraints are satisfied at evaluation time, given that they are satisfied at training time. To improve generalization performance, we frame the problem as a two-player game where one player optimizes the model parameters on a training dataset, and the other player enforces the constraints on an independent validation dataset. We build on recent work in two-player constrained optimization to show that if one uses this two-dataset approach, then constraint generalization can be significantly improved. As we illustrate experimentally, this approach works not only in theory, but also in practice.

Subfields: Artificial Intelligence, Safety Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

Citations in the corpus

Beyond Individual and Group Fairness (2020, background): “These publications either reduce the problem to that of cost-sensitive classification [Agarwal et al., 2018, Dwork et al., 2018] or replace the non-convex constraints by convex proxies and next optimize them via external or swap regret minimization algorithms [Cotter et al., 2018b,a].”
Optimization with Non-Differentiable Constraints with Applications to Fairness, Recall, Churn, and Other Goals (2018, methodology): “…(1972); Groeneboom and Jongbloed (2014); Gupta et al. (2016); Canini et al. (2016); Luss and Rosset (2017); You et al. (2017); Bonakdarpour et al. (2018)), or other shapes (e.g. Chetverikov et al. (2018); Pya and Wood (2015); Chen and Samworth (2016); Gupta et al. (2018); Cotter et al. (2019b)).”
Noise-tolerant fair classification (2019, background): “, 2018), alternating minimisation (Zafar et al., 2017b; Cotter et al., 2018), or linearisation (Hardt et al.”
Two-Player Games for Efficient Non-Convex Constrained Optimization (2018, background): “Frank-Wolfe, as in Hazan and Kale (2012); Jaggi (2013); Garber and Hazan (2013)).”
Fairness Violations and Mitigation under Covariate Shift (2019, background): “In [13], authors propose algorithms for generalisation of fairness constraints but to an i.”
Fairness in Machine Learning (2020, methodology): “Some methods transform the constrained optimization problem via the method of Lagrange multipliers [3, 15, 34, 37, 38, 59, 97, 135, 183, 185] or add penalties to the objective [5, 14, 22, 44, 46, 56, 58, 62, 73, 74, 75, 79, 82, 87, 88, 89, 93, 94, 95, 98, 104, 115, 119, 123, 133, 134, 135, 138, 139,”
Identifying and Correcting Label Bias in Machine Learning (2019, methodology): “In the non-convex case, there is work which frames the constrained optimization problem as a two-player game (Kearns et al., 2017; Agar-wal et al., 2018; Cotter et al., 2018b) .”
Lagrangian Duality for Constrained Deep Learning (2020, background): “[11] propose a two-players game approach in which one player optimizes the model parameters on the training data and the other player optimizes the constraints on a validation set.”
Pairwise Fairness for Ranking and Regression (2019, methodology): “…Many statistical fairness metrics for binary classification can be written in terms of rate constraints, that is, constraints on the classifier’s positive (or negative) prediction rate for different groups (Goh et al. 2016; Narasimhan 2018; Cotter, Jiang, and Sridharan 2019; Cotter et al. 2019).”