Embedding differential privacy in decision tree algorithm with different depths

Bai, Xuanyu; Yao, Jianguo; Yuan, Mingxuan; Deng, Ke; Xie, Xike; Guan, Haibing

doi:10.1007/s11432-016-0442-1

Cited by 11 publications

(8 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For optimization reasons, several works modify the quality metric used by the learning algorithm [5,17,34,38,53,54,112,114,122,149,178,183] (see §6). Overall in the literature, we observe a wide range of combinations on tree-types, tasks, data, and algorithms, as illustrated CART [34] CART-like [5] C4.5-like [168] CRT [116] ExRT [119] ExRT [62] CART [15,17,71,85,96,110,152,154,158,181] C4.5 [41] CART [172,192] CART-like [156] CRT [68,171] ID3/C4.5 [112] C45 [7,8,27] CART [9] C4.5 [16,26,79,183] CART [95,108,177] CRT [84] ID3/CRT [20] CRF …”

Section: Learning Algorithmmentioning

confidence: 99%

“…The main idea is to inject noise during key learning parts, e.g., for selecting the best feature [68], counting class counts at the leaves [34,84], or computing gain queries for each feature [62]. More recent works aim to find tighter sensitivity bounds for the training queries or new ways to embed DP [16]. Other approaches relax the learning algorithm by replacing information gain with more DP-friendly metrics, e.g., Gini [62] or Max [114].…”

Section: Differential Privacy Based Solutionsmentioning

confidence: 99%

“…However, this is circumvented by training each tree on an independent subset of the training data and applying the parallel composition theorem [59,183]. Overall, works based on the central model consider reasonable privacy budgets (i.e., ε∈[0.1; 1.0]) and some even experiment with very low budgets (e.g., ε=0.01 for [16,20,59,114,137]). We note that the privacy budget configuration directly affects the model's accuracy; Fletcher et al report accuracy drops of more than 20% when the budget is reduced from 2.0 to 0.2 to obtain stronger privacy guarantees [60].…”

Section: Differential Privacy Based Solutionsmentioning

confidence: 99%

“…We find that most works ensure the confidentiality of the raw training data, but do not consider the leakage that might occur from computations required for collaborative tree induction, e.g., comparison operations are performed on cleartext data to abide with HE limitations. To this end, we design a framework that analyzes which information is leaked during privacy-preserving collab- N/2 [48,179,190] N −1 [70,80,153] N −2 [32] N −u [167,168] N −τ [17, 37, 41, 42, 55, 56, 71, 72, 95, 96, 100, 104, 111, 121, 122, 149, 156, 157, 160, 166, 181, 182, 186, 187] [5] N/2 [101,102] TH TH [16,19,20,34,59,60,62,68,79,83,84,105,114,137,140,146,171,177,183,188,192] orative tree-based model induction, enabling us to systematize the literature on that aspect. Although several works acknowledge this leakage and even provide an analysis of their solution [32,55,70,75,118,167,168,179], they do not do so in a systematic way.…”

Section: Leakage Analysismentioning

confidence: 99%

See 3 more Smart Citations

SoK: Privacy-Preserving Collaborative Tree-based Model Learning

Chatel

Pyrgelis

Troncoso-Pastoriza

et al. 2021

Proceedings on Privacy Enhancing Technologies

View full text Add to dashboard Cite

Tree-based models are among the most efficient machine learning techniques for data mining nowadays due to their accuracy, interpretability, and simplicity. The recent orthogonal needs for more data and privacy protection call for collaborative privacy-preserving solutions. In this work, we survey the literature on distributed and privacy-preserving training of tree-based models and we systematize its knowledge based on four axes: the learning algorithm, the collaborative model, the protection mechanism, and the threat model. We use this to identify the strengths and limitations of these works and provide for the first time a framework analyzing the information leakage occurring in distributed tree-based model learning.

show abstract

Section: Learning Algorithmmentioning

confidence: 99%

Section: Differential Privacy Based Solutionsmentioning

confidence: 99%

Section: Differential Privacy Based Solutionsmentioning

confidence: 99%

Section: Leakage Analysismentioning

confidence: 99%

See 2 more Smart Citations

SoK: Privacy-Preserving Collaborative Tree-based Model Learning

Chatel

Pyrgelis

Troncoso-Pastoriza

et al. 2021

Proceedings on Privacy Enhancing Technologies

View full text Add to dashboard Cite

show abstract

“…For a fixed accuracy, which techniques provide better privacy?). Existing research relating to DP include mechanism design, 14 applications, 15–18 learning, 19,20 etc.…”

Section: Related Literaturementioning

confidence: 99%

Heterogeneous differential privacy for vertically partitioned databases

Xia

Zhu

Ding

et al. 2019

Concurrency and Computation

View full text Add to dashboard Cite

Summary Existing privacy‐preserving approaches are generally designed to provide privacy guarantee for individual data in a database, which reduces the utility of the database for data analysis. In this paper, we propose a novel differential privacy mechanism to preserve the heterogeneous privacy of a vertically partitioned database based on attributes. We first present the concept of privacy label, which characterizes the privacy information of the database and is instantiated by the classification. Then, we use an information‐based method to systematically explore the dependencies between all attributes and the privacy label. We finally assign privacy weights to every attribute and design a heterogeneous mechanism according to the basic Laplace mechanism. Evaluations using real datasets demonstrate that the proposed mechanism achieves a balanced privacy and utility.

show abstract

Differentially private tree-based redescription mining

Mihelčić

Miettinen

2023

Data Min Knowl Disc

View full text Add to dashboard Cite

Differential privacy provides a strong form of privacy and allows preserving most of the original characteristics of the dataset. Utilizing these benefits requires one to design specific differentially private data analysis algorithms. In this work, we present three tree-based algorithms for mining redescriptions while preserving differential privacy. Redescription mining is an exploratory data analysis method for finding connections between two views over the same entities, such as phenotypes and genotypes of medical patients, for example. It has applications in many fields, including some, like health care informatics, where privacy-preserving access to data is desired. Our algorithms are the first tree-based differentially private redescription mining algorithms, and we show via experiments that, despite the inherent noise in differential privacy, it can return trustworthy results even in smaller datasets where noise typically has a stronger effect.

show abstract

Embedding differential privacy in decision tree algorithm with different depths

Cited by 11 publications

References 6 publications

SoK: Privacy-Preserving Collaborative Tree-based Model Learning

SoK: Privacy-Preserving Collaborative Tree-based Model Learning

Heterogeneous differential privacy for vertically partitioned databases

Differentially private tree-based redescription mining

Contact Info

Product

Resources

About