Robustness of Maximal α-Leakage to Side Information

Liao, Jiachun; Sankar, Lalitha; Kosut, Oliver; Calmon, Flávio P.

doi:10.1109/isit.2019.8849769

Cited by 13 publications

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are estimation challenges for information-theoretic measures (see [65,66] and the references therein). Designing estimators building upon techniques proposed in [25,27,66,67] is an interesting direction of research. In [68], a method for estimating the unique information for continuous distributions is proposed.…”

Section: Estimation Of Pid Measuresmentioning

confidence: 99%

“…Classical information-theoretic measures, such as mutual information, are widely used [9,[25][26][27][28] in fairness to quantify the disparity (dependence) with respect to the sensitive attribute (Z) in the output of a model ( Ŷ). In several situations, however, only identifying disparity in the final output of a model (e.g., quantifying I(Z; Ŷ)) is not enough.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Review of Partial Information Decomposition in Algorithmic Fairness and Explainability

Dutta

Faisal

2023

Entropy

View full text Add to dashboard Cite

Partial Information Decomposition (PID) is a body of work within information theory that allows one to quantify the information that several random variables provide about another random variable, either individually (unique information), redundantly (shared information), or only jointly (synergistic information). This review article aims to provide a survey of some recent and emerging applications of partial information decomposition in algorithmic fairness and explainability, which are of immense importance given the growing use of machine learning in high-stakes applications. For instance, PID, in conjunction with causality, has enabled the disentanglement of the non-exempt disparity which is the part of the overall disparity that is not due to critical job necessities. Similarly, in federated learning, PID has enabled the quantification of tradeoffs between local and global disparities. We introduce a taxonomy that highlights the role of PID in algorithmic fairness and explainability in three main avenues: (i) Quantifying the legally non-exempt disparity for auditing or training; (ii) Explaining contributions of various features or data points; and (iii) Formalizing tradeoffs among different disparities in federated learning. Lastly, we also review techniques for the estimation of PID measures, as well as discuss some challenges and future directions.

show abstract

Section: Estimation Of Pid Measuresmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Review of Partial Information Decomposition in Algorithmic Fairness and Explainability

Dutta

Faisal

2023

Entropy

View full text Add to dashboard Cite

show abstract

“…The idea of Algorithms 1 and 2 in this paper is analogous to the agglomerative pairwise merge algorithms in [33], [35]. Similar to L 0 (S → X), the privacy in information theory is measured in terms of the logarithm of the fraction between the prior and posterior statistical uncertainty on S. For example, the average inference loss defined as H(S) − H(S| X), or the mutual information I(S; X), is extended to the worst case H(S) − max x H(S|x) [7] and the α-leakage [9], [36], where the latter is a tunable measure of the mutual information.…”

Section: Notes On Stochastic Information-theoretic Privacy and Differ...mentioning

confidence: 99%

Developing Non-Stochastic Privacy-Preserving Policies Using Agglomerative Clustering

Ding¹,

Farokhi²

2019

Preprint

View full text Add to dashboard Cite

We consider a non-stochastic privacy-preserving problem in which an adversary aims to infer sensitive information S from publicly accessible data X without using statistics. We consider the problem of generating and releasing a quantization X of X to minimize the privacy leakage of S to X while maintaining a certain level of utility (or, inversely, the quantization loss). The variables S and X are treated as bounded and non-probabilistic, but are otherwise general. We consider two existing non-stochastic privacy measures, namely the maximum uncertainty reduction L0(S → X) and the refined information I * (S; X) (also called the maximin information) of S. For each privacy measure, we propose a corresponding agglomerative clustering algorithm that converges to a locally optimal quantization solution X by iteratively merging elements in the alphabet of X. To instantiate the solution to this problem, we consider two specific utility measures, the worst-case resolution of X by observing X and the maximal distortion of the released data X. We show that the value of the maximin information I * (S; X) can be determined by dividing the confusability graph into connected subgraphs. Hence, I * (S; X) can be reduced by merging nodes connecting subgraphs. The relation to the probabilistic information-theoretic privacy is also studied by noting that the Gács-Körner common information is the stochastic version of I * and indicates the attainability of statistical indistinguishability.

show abstract

“…Intuitively, in this setup, observations only convey the unique information contributed by the unknown data entry since all other entries are already known to the adversary. To quantify this entrywise information leakage, we propose a conditional form of maximal leakage, namely the pointwise conditional maximal leakage, which is also a special case of the event-conditional Sibson mutual information introduced in [13]. Then, by allowing the unknown entry to be any of the entries in the dataset, we can derive upper bounds on the entrywise information leakage, and provide meaningful worst-case privacy guarantees.…”

Section: Introductionmentioning

confidence: 99%

Quantifying Membership Privacy via Information Leakage

Saeidian

Cervia

Oechtering

et al. 2021

IEEE Trans.Inform.Forensic Secur.

View full text Add to dashboard Cite

Machine learning models are known to memorize the unique properties of individual data points in a training set. This memorization capability can be exploited by several types of attacks to infer information about the training data, most notably, membership inference attacks. In this paper, we propose an approach based on information leakage for guaranteeing membership privacy. Specifically, we propose to use a conditional form of the notion of maximal leakage to quantify the information leaking about individual data entries in a dataset, i.e., the entrywise information leakage. We apply our privacy analysis to the Private Aggregation of Teacher Ensembles (PATE) framework for privacy-preserving classification of sensitive data and prove that the entrywise information leakage of its aggregation mechanism is Schur-concave when the injected noise has a logconcave probability density. The Schur-concavity of this leakage implies that increased consensus among teachers in labeling a query reduces its associated privacy cost. Finally, we derive upper bounds on the entrywise information leakage when the aggregation mechanism uses Laplace distributed noise.

show abstract

Robustness of Maximal α-Leakage to Side Information

Cited by 13 publications

References 13 publications

A Review of Partial Information Decomposition in Algorithmic Fairness and Explainability

A Review of Partial Information Decomposition in Algorithmic Fairness and Explainability

Developing Non-Stochastic Privacy-Preserving Policies Using Agglomerative Clustering

Quantifying Membership Privacy via Information Leakage

Contact Info

Product

Resources

About