Achieving Non-Discrimination in Data Release

By Lu Zhang, Yongkai Wu, Xintao Wu. 2016

74 citations recorded by the source corpus.

Discrimination discovery and prevention/removal are increasingly important tasks in data mining. Discrimination discovery aims to unveil discriminatory practices on the protected attribute (e.g., gender) by analyzing the dataset of historical decision records, and discrimination prevention aims to remove discrimination by modifying the biased data before conducting predictive analysis. In this paper, we show that the key to discrimination discovery and prevention is to find the meaningful partitions that can be used to provide quantitative evidences for the judgment of discrimination. With the support of the causal graph, we present a graphical condition for identifying a meaningful partition. Based on that, we develop a simple criterion for the claim of non-discrimination, and propose discrimination removal algorithms which accurately remove discrimination while retaining good data utility. Experiments using real datasets show the effectiveness of our approaches.

Subfields: Artificial Intelligence, Safety Research, Privacy & Security

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

In Privacy & Security

Citations in the corpus

A Survey on Bias and Fairness in Machine Learning (2019, methodology): “Causal models and graphs were also used in some work to detect direct discrimination in the data along with its prevention technique that modifies the data such that the predictions would be absent from direct discrimination [134].”
Causal Modeling-Based Discrimination Discovery and Removal: Criteria, Bounds, and Algorithms (2019, methodology): “Studies in [7], [8], [9] are built on causal modeling and the associated causal graph, but cannot deal with indirect discrimination.”
FairGAN: Fairness-aware Generative Adversarial Networks (2018, background): “Research in [18] proposed a causal graph based approach that removes discrimination based on the block set and ensures that there is no discrimination in any meaningful partition.”
Causal Discrimination Discovery Through Propensity Score Analysis (2016, methodology): “ndirect discrimination), and of common causes (spurious discrimination) on the decision attribute. In this paper, instead, we deal only with direct discrimination. Finally, some approaches (see e.g., [37]) are extended to tackle training data sanitization. While this is out of the scope of this pape”
Say No to the Discrimination: Learning Fair Graph Neural Networks with Limited Sensitive Attribute Information (2020, background): “They could reduce the bias by modifying the training data through correcting labels [22, 56], revising attributes of data [14, 23], generating non-discriminatory labeled data [41, 48, 49], and obtaining fair data representations [3, 9, 13, 29–31, 33, 42, 54].”
PC-Fairness: A Unified Framework for Measuring Causality-based Fairness (2019, unknown): “Table 1: Connection between previous fairness notions and PC fairness Description References Relating to PC fairness Total effect [19, 16] O = ∅ and π = Π (System) Direct discrimination [19, 7, 16] O = ∅ or {S} and π = πd = {S → Ŷ } (System) Indirect discrimination [19, 7, 16] O = ∅ or {S} and π = π”
Anti-discrimination learning: a causal modeling-based framework (2017, methodology): “The work [38] considers group-level direct discrimination as the discrimination measured on subgroups produced by partitioning the dataset.”
Achieving non-discrimination in prediction (2017, methodology): “, 2011] and the CBN-Based Removal [Zhang et al., 2016], and the methods that modify the data other than the label, including the Preferential Sampling [Kamiran and Calders, 2012; Žliobaitė et al.”
Fairness in Machine Learning (2020, methodology): “Pre-Processing methods consist in transforming the training data to remove undesired biases [16, 21, 24, 28, 47, 51, 52, 53, 61, 64, 65, 66, 67, 86, 90, 91, 92, 112, 114, 116, 118, 121, 122, 127, 126, 164, 174, 188, 190, 191, 193].”