國立虎尾科技大學 |

Privacy and Efficiency in Personalized Decision-Making and Recommendation.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Privacy and Efficiency in Personalized Decision-Making and Recommendation./
Author:	Carranza, Aldo Gael.
Description:	1 online resource (218 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-04, Section: B.
Contained By:	Dissertations Abstracts International85-04B.
Subject:	Decision making. -
Online resource:	click for full text (PQDT)
ISBN:	9798380469920

Privacy and Efficiency in Personalized Decision-Making and Recommendation.
Carranza, Aldo Gael.

Privacy and Efficiency in Personalized Decision-Making and Recommendation. - 1 online resource (218 pages)

Source: Dissertations Abstracts International, Volume: 85-04, Section: B.

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

Includes bibliographical references

In the current digital era, marked by the ubiquity of individual-level data and sophisticated artificial intelligence systems highly capable of exploiting data heterogeneity, data-driven personalized decision-making and recommendation systems have become prevalent in providing customized services and experiences to individuals. These bespoke systems, however, continue to present considerable challenges regarding their deployment in adaptive, heterogeneous, and privacy-sensitive settings. This dissertation presents three research projects that delve into some of these critical issues, offering insights and novel solutions aimed at enhancing the privacy and efficiency of personalized decision-making and recommendation systems.Chapter 1 of this dissertation investigates the challenges of model learning in contextual bandits for adaptive decision-making and presents a method to improve data efficiency and robustness to model misspecification in this online setting. Contextual bandit algorithms often estimate reward models to inform decision-making. However, true rewards can contain action-independent redundancies that are not relevant for decision-making. We show it is more data-efficient to estimate any function that explains the reward differences between actions, that is, the treatment effects. Motivated by this observation, building on recent work on oracle-based bandit algorithms, we provide a universal reduction of contextual bandits to general-purpose heterogeneous treatment effect estimation, and we design a simple and computationally efficient algorithm based on this reduction. Our theoretical and experimental results demonstrate that heterogeneous treatment effect estimation in contextual bandits offers practical advantages over reward estimation, including more efficient model estimation and greater robustness to model misspecification.In Chapter 2, we consider heterogeneous data adaptation and privacy in decision-making informed by historical observational data. We consider the problem of learning personalized decision policies on observational bandit feedback data from heterogeneous data sources. Moreover, we examine this problem in the federated setting where a central server aims to learn a policy on the data distributed across the heterogeneous sources without exchanging their raw data due to privacy concerns. We present a federated policy learning algorithm based on aggregation of local policies trained with doubly robust offline policy evaluation and learning strategies. We provide a novel regret analysis for our approach that establishes a finite-sample upper bound on a notion of global regret across a distribution of clients. In addition, for any individual client, we establish a corresponding local regret upper bound characterized by the presence of distribution shift relative to all other clients. We support our theoretical findings with experimental results. Our analysis and experiments provide insights into the value of heterogeneous client participation in federation for policy learning in heterogeneous settings.Lastly, in Chapter 3, we pivot slightly from the focus of the first two chapters on decision-making systems with online and offline policy learning methods to investigating data privacy in recommender systems. We propose a novel approach for developing privacy-preserving large-scale recommender systems using differentially private (DP) large language models (LLMs) which overcomes certain challenges and limitations in DP training these complex systems. This method is particularly well suited for the emerging area of LLM-based recommender systems, but can be readily employed for any recommender systems that process representations of natural language inputs. Our approach involves using DP training methods to fine-tune a publicly pre-trained LLM on a query generation task.

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

Mode of access: World Wide Web

ISBN: 9798380469920Subjects--Topical Terms:

528319
Decision making.
Index Terms--Genre/Form:

554714
Electronic books.

Privacy and Efficiency in Personalized Decision-Making and Recommendation.
LDR:05094ntm a22003377 4500 001 1148710
005 20240930100123.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798380469920
035 $a (MiAaPQ)AAI30615162
035 $a (MiAaPQ)STANFORDky391xq2715
035 $a AAI30615162
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Carranza, Aldo Gael. $3 1474749
245 1 0 $a Privacy and Efficiency in Personalized Decision-Making and Recommendation.
264 0 $c 2023
300 $a 1 online resource (218 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: 85-04, Section: B.
500 $a Advisor: Athey, Susan.
502 $a Thesis (Ph.D.)--Stanford University, 2023.
504 $a Includes bibliographical references
520 $a In the current digital era, marked by the ubiquity of individual-level data and sophisticated artificial intelligence systems highly capable of exploiting data heterogeneity, data-driven personalized decision-making and recommendation systems have become prevalent in providing customized services and experiences to individuals. These bespoke systems, however, continue to present considerable challenges regarding their deployment in adaptive, heterogeneous, and privacy-sensitive settings. This dissertation presents three research projects that delve into some of these critical issues, offering insights and novel solutions aimed at enhancing the privacy and efficiency of personalized decision-making and recommendation systems.Chapter 1 of this dissertation investigates the challenges of model learning in contextual bandits for adaptive decision-making and presents a method to improve data efficiency and robustness to model misspecification in this online setting. Contextual bandit algorithms often estimate reward models to inform decision-making. However, true rewards can contain action-independent redundancies that are not relevant for decision-making. We show it is more data-efficient to estimate any function that explains the reward differences between actions, that is, the treatment effects. Motivated by this observation, building on recent work on oracle-based bandit algorithms, we provide a universal reduction of contextual bandits to general-purpose heterogeneous treatment effect estimation, and we design a simple and computationally efficient algorithm based on this reduction. Our theoretical and experimental results demonstrate that heterogeneous treatment effect estimation in contextual bandits offers practical advantages over reward estimation, including more efficient model estimation and greater robustness to model misspecification.In Chapter 2, we consider heterogeneous data adaptation and privacy in decision-making informed by historical observational data. We consider the problem of learning personalized decision policies on observational bandit feedback data from heterogeneous data sources. Moreover, we examine this problem in the federated setting where a central server aims to learn a policy on the data distributed across the heterogeneous sources without exchanging their raw data due to privacy concerns. We present a federated policy learning algorithm based on aggregation of local policies trained with doubly robust offline policy evaluation and learning strategies. We provide a novel regret analysis for our approach that establishes a finite-sample upper bound on a notion of global regret across a distribution of clients. In addition, for any individual client, we establish a corresponding local regret upper bound characterized by the presence of distribution shift relative to all other clients. We support our theoretical findings with experimental results. Our analysis and experiments provide insights into the value of heterogeneous client participation in federation for policy learning in heterogeneous settings.Lastly, in Chapter 3, we pivot slightly from the focus of the first two chapters on decision-making systems with online and offline policy learning methods to investigating data privacy in recommender systems. We propose a novel approach for developing privacy-preserving large-scale recommender systems using differentially private (DP) large language models (LLMs) which overcomes certain challenges and limitations in DP training these complex systems. This method is particularly well suited for the emerging area of LLM-based recommender systems, but can be readily employed for any recommender systems that process representations of natural language inputs. Our approach involves using DP training methods to fine-tune a publicly pre-trained LLM on a query generation task.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Decision making. $3 528319
650 4 $a Computer science. $3 573171
650 4 $a Recommender systems. $3 1372552
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Stanford University. $3 1184533
773 0 $t Dissertations Abstracts International $g 85-04B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30615162 $z click for full text (PQDT)