A Variational Approach to Privacy and Fairness

Rodríguez-Gálvez, Borja; Thobaben, Ragnar; Skoglund, Mikael

doi:10.1109/itw48936.2021.9611429

Cited by 15 publications

(21 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As shown in Table I, our method outperforms all other methods, showing improvements in accuracy and fairness across all metrics. Usual caution should be exercised in interpretations since -despite our aligning with data partitioning in [25] -other variations may exist with [6], [9], [25] due to nondeterminism, parameter setting or other factors.…”

Section: Resultsmentioning

confidence: 99%

“…for α > 1. The first inequality follows from the non-negativity of Kullback-Leibler (KL) divergence, similar to [2], [6], [9]. For the final step, we take the Rényi divergence D α (• •) of order α (e.g., see [20]), rather than the KL divergence as typically done in the literature, where…”

Section: Variational Boundsmentioning

confidence: 99%

“…2 For simplicity and by the continuity property of Dα, we define its extended orders at α = 1 and α = 0 [20] as D 1 (P Q) := D KL (P Q) and D 0 (P Q) := lim α→0 Dα(P Q) = − log Q(x : P (x) > 0), which is equal to 0 when P and Q share a common support. We note that depending on the choice of α, β 1 , and β 2 , from our method we can recover both the IB [9] and conditional fairness bottleneck (CFB) [6] schemes to which we compare our results. Letting α = 1 and β 2 = 0 corresponds to IB, while setting α = 1 and β 1 = 0 corresponds to CFB.…”

Section: Computing the Boundsmentioning

confidence: 99%

“…2) Measures of Fairness: We measure this in multiple ways: a) using the gap in accuracy (denoted acc gap ) between favored and protected subpopulations; and b) reporting the minimum accuracy across subpopulations (denoted as acc min ), which is based on the Rawlsian principle of achieving fairness by maximizing acc min [24]. Also echoing [6], we measure: c) the adherence to the demographic parity via its gap dp gap ; and adherence to d) equal odds via its gap eqodds gap . The latter two metrics are respectively given by:…”

Section: B Metricsmentioning

confidence: 99%

“…Learning fair representations can be done both by adversarial methods, such as [3]- [5], and through variational approaches such as [6], [7] that maximize and minimize different mutual information terms to ensure the representation is both expressive and fair. These approaches are related to the information bottleneck (IB) method [8] and its variational approximations [9]- [11] which aim to find a representation This work was supported in part by NSERC of Canada.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Renyi Fair Information Bottleneck for Image Classification

Gronowski¹,

Paul²,

Alajaji³

et al. 2022

Preprint

View full text Add to dashboard Cite

We develop a novel method for ensuring fairness in machine learning which we term as the Rényi Fair Information Bottleneck (RFIB). We consider two different fairness constraints -demographic parity and equalized odds -for learning fair representations and derive a loss function via a variational approach that uses Rényi 's divergence with its tunable parameter α and that takes into account the triple constraints of utility, fairness, and compactness of representation. We then evaluate the performance of our method for image classification using the EyePACS medical imaging dataset, showing it outperforms competing state of the art techniques with performance measured using a variety of compound utility/fairness metrics, including accuracy gap and Rawls' minimal accuracy.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Variational Boundsmentioning

confidence: 99%

Section: Computing the Boundsmentioning

confidence: 99%

Section: B Metricsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Renyi Fair Information Bottleneck for Image Classification

Gronowski¹,

Paul²,

Alajaji³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

Ensuring generalized fairness in batch classification

Pal,

Pokhriyal,

Sikdar

et al. 2023

Sci Rep

View full text Add to dashboard Cite

In this paper, we consider the problem of batch classification and propose a novel framework for achieving fairness in such settings. The problem of batch classification involves selection of a set of individuals, often encountered in real-world scenarios such as job recruitment, college admissions etc. This is in contrast to a typical classification problem, where each candidate in the test set is considered separately and independently. In such scenarios, achieving the same acceptance rate (i.e., probability of the classifier assigning positive class) for each group (membership determined by the value of sensitive attributes such as gender, race etc.) is often not desirable, and the regulatory body specifies a different acceptance rate for each group. The existing fairness enhancing methods do not allow for such specifications and hence are unsuited for such scenarios. In this paper, we define a configuration model whereby the acceptance rate of each group can be regulated and further introduce a novel batch-wise fairness post-processing framework using the classifier confidence-scores. We deploy our framework across four real-world datasets and two popular notions of fairness, namely demographic parity and equalized odds. In addition to consistent performance improvements over the competing baselines, the proposed framework allows flexibility and significant speed-up. It can also seamlessly incorporate multiple overlapping sensitive attributes. To further demonstrate the generalizability of our framework, we deploy it to the problem of fair gerrymandering where it achieves a better fairness-accuracy trade-off than the existing baseline method.

show abstract