2013
DOI: 10.48550/arxiv.1309.3958
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Utilizing Noise Addition for Data Privacy, an Overview

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
13
0
2

Year Published

2013
2013
2024
2024

Publication Types

Select...
4
3

Relationship

0
7

Authors

Journals

citations
Cited by 16 publications
(15 citation statements)
references
References 26 publications
0
13
0
2
Order By: Relevance
“…The authors in [29] introduced a model for local obfuscation of probability distribution by probability perturbation, which perturbed each single "point" datum by adding controlled probabilistic noise before sending it out to a data collector. The authors in [30] studied noise addition as a data privacy providing technique.…”
Section: A Additivementioning
confidence: 99%
See 1 more Smart Citation
“…The authors in [29] introduced a model for local obfuscation of probability distribution by probability perturbation, which perturbed each single "point" datum by adding controlled probabilistic noise before sending it out to a data collector. The authors in [30] studied noise addition as a data privacy providing technique.…”
Section: A Additivementioning
confidence: 99%
“…Obfuscation technique Smart grid Comments [16] Additive Yes Used Laplace distribution [17] Additive Yes Used Gaussian distribution [18] Additive Yes Used Gaussian distribution [23] Additive Yes Used Laplace distribution [24] Additive Yes Used uniform distribution [25] Miscellaneous Yes Used GMM [27] Additive Yes Used correlated Gaussian noise [28] Additive Yes Used Laplace distribution [29] Additive No Used controlled probabilistic noise [30] Additive No Studied noise addition [31] Multiplicative No Used truncated triangular distribution [32] Multiplicative No Used orthogonal matrices [33] Multiplicative No Showed that if original and perturbed data are highly correlated, a malicious entity can recover the original data [34] Multiplicative No Stated that multiplicative noise has the advantage of perturbation size directly proportional to the original data value [35] Multiplicative No Evaluated linear and non-linear schemes [36] Additive + Yes Showed that data hiding and additive noise could hinder consumer attributes idetification [37] Additive + Yes Used uniform noise followed by homomorphic encryption [38] Additive + No Either added random noise or augmented the data with fake noisy samples [39] Additive + No Studied obfuscation and anonymization, considered privacy but not utility [40] Miscellaneous Yes Used a sparse dictionary [41] Miscellaneous No Used GANs for privacy-reservation of mobile datasets [42] Miscellaneous Yes Studied linear and data shuffling masking methods that required knowledge of the whole data set [43] Miscellaneous Yes Replaced high-risk data with alternative data data by using linear regression for a large number of consumers.…”
Section: Ref Nomentioning
confidence: 99%
“…Indeed, complex noise transformations severely alter the semantics of data resulting in a significant utility loss, as described in [11], [14]. On the other hand, simple noise distortion techniques can be reverted to obtain the original microdata set, as demonstrated by [11], [15], [16].…”
Section: Anonymization Techniquesmentioning
confidence: 99%
“…Perturbation distorts the data by adding noise, aggregating values, swapping values, or generating synthesized data based on statistical properties of the original [25]. Since these records do not correspond to real-world owners, the attacker cannot perform the same sensitive linkage attacks found with generalization.…”
Section: Noisementioning
confidence: 99%
“…This transformation of sensitive values works by adding or multiplying a stochastic or randomized value. The stochastic value is chosen from the "normal" distribution with zero mean and a diminutive standard deviation [16], [25]. Additive noise, also known as white noise, is adding a stochastic value to the original sensitive value which then replaces the original for publication, and this is known as the the transformed data point, which preservers the mean and co-variance [12] of the uniform distributed noise.…”
Section: Noisementioning
confidence: 99%