國立虎尾科技大學 |

Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning./
作者:	Zakynthinou, Lydia.
面頁冊數:	1 online resource (241 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
Contained By:	Dissertations Abstracts International85-02B.
標題:	Computer science. -
電子資源:	click for full text (PQDT)
ISBN:	9798380114950

Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning.
Zakynthinou, Lydia.

Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning. - 1 online resource (241 pages)

Source: Dissertations Abstracts International, Volume: 85-02, Section: B.

Thesis (Ph.D.)--Northeastern University, 2023.

Includes bibliographical references

Machine learning algorithms aim to learn useful models for the population, only having access to a training dataset as its proxy. However, they may overly tailor their output to the training dataset, which leads to the following two types of unwanted behavior: 1. they may leak too much information about specific datapoints included in the training dataset, violating the privacy of the individuals who participate in it, or 2. they may "overfit" their training dataset, that is, their empirical performance on the training dataset does not generalize well to their true performance on the population. This thesis contributes tools for preventing or reasoning about these two unwanted behaviors. 1. We contribute privacy-preserving algorithms for two fundamental tasks: learning linear classifiers for separable data and learning the mean of unbounded Gaussian data. Our algorithms have optimal error and satisfy differential privacy (DP), a rigorous mathematical condition that prevents privacy leakage. Moreover, we propose a systematic framework that allows us to transform algorithms that are robust to training-data corruptions into DP algorithms, thus retrieving near-optimal DP algorithms for a variety of statistical tasks. 2. We propose a framework based on a new information-theoretic notion, termed conditional mutual information (CMI), that allows us to reason about the generalization guarantees of machine learning algorithms. Our framework unifies several seemingly-incomparable, existing approaches, in the sense that they all imply that an algorithm satisfies our new notion of stability and, in turn, the latter implies that it generalizes well to unseen data. The unifying nature of our framework makes it a versatile tool in proving generalization guarantees of machine learning algorithms.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798380114950Subjects--Topical Terms:

573171
Computer science.
Subjects--Index Terms:

Algorithmic stabilityIndex Terms--Genre/Form:

554714
Electronic books.

Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning.
LDR:03232ntm a22003977 4500 001 1148714
005 20240930100125.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798380114950
035 $a (MiAaPQ)AAI30635049
035 $a AAI30635049
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Zakynthinou, Lydia. $3 1474753
245 1 0 $a Algorithms and Frameworks for Preventing Privacy Leakage and Overfitting in Machine Learning.
264 0 $c 2023
300 $a 1 online resource (241 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
500 $a Advisor: Ullman, Jonathan;Nguyễn, Huy Le.
502 $a Thesis (Ph.D.)--Northeastern University, 2023.
504 $a Includes bibliographical references
520 $a Machine learning algorithms aim to learn useful models for the population, only having access to a training dataset as its proxy. However, they may overly tailor their output to the training dataset, which leads to the following two types of unwanted behavior: 1. they may leak too much information about specific datapoints included in the training dataset, violating the privacy of the individuals who participate in it, or 2. they may "overfit" their training dataset, that is, their empirical performance on the training dataset does not generalize well to their true performance on the population. This thesis contributes tools for preventing or reasoning about these two unwanted behaviors. 1. We contribute privacy-preserving algorithms for two fundamental tasks: learning linear classifiers for separable data and learning the mean of unbounded Gaussian data. Our algorithms have optimal error and satisfy differential privacy (DP), a rigorous mathematical condition that prevents privacy leakage. Moreover, we propose a systematic framework that allows us to transform algorithms that are robust to training-data corruptions into DP algorithms, thus retrieving near-optimal DP algorithms for a variety of statistical tasks. 2. We propose a framework based on a new information-theoretic notion, termed conditional mutual information (CMI), that allows us to reason about the generalization guarantees of machine learning algorithms. Our framework unifies several seemingly-incomparable, existing approaches, in the sense that they all imply that an algorithm satisfies our new notion of stability and, in turn, the latter implies that it generalizes well to unseen data. The unifying nature of our framework makes it a versatile tool in proving generalization guarantees of machine learning algorithms.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 573171
650 4 $a Statistics. $3 556824
653 $a Algorithmic stability
653 $a Differential privacy
653 $a Generalization
653 $a Reliable machine learning
653 $a Robustness
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0463
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Northeastern University. $b Computer Science. $3 1464678
773 0 $t Dissertations Abstracts International $g 85-02B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30635049 $z click for full text (PQDT)