서브메뉴
검색
MCMC With Substitutions and Multi-Armed Bandits With Covariates: Theory and Applications
MCMC With Substitutions and Multi-Armed Bandits With Covariates: Theory and Applications
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104745
- ISBN
- 9798290652245
- DDC
- 500
- 저자명
- Xu, Huanzhong.
- 서명/저자
- MCMC With Substitutions and Multi-Armed Bandits With Covariates: Theory and Applications
- 발행사항
- [Sl] : Stanford University, 2023
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- 형태사항
- 94 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-03, Section: B.
- 주기사항
- Advisor: Mancilla, Jose Blanchet.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2023.
- 초록/해제
- 요약The thesis covers three areas of study: Markov Chain Monte Carlo, multi-armed bandits, and reliability of object detection models. Each part begins with a review of relevant literature and a discussion of the motivation behind our novel algorithm. We then present our main theoretical findings, accompanied by proofs detailed in the appendices. Finally, we showcase the results of our simulation studies and real-world applications of our algorithms.The first part is in the area of multi-armed bandits, and we introduce a new approach to nonparametric multi-armed bandit theory involving both the bandit and the covariate processes. The extension to bandit processes with a non-denumerable set of arms is also discussed. The approach we develop herein can be readily extended to continuous-time processes by using ε-greedy randomization and arm elimination instead of dynamic allocation indices. It also carries out a stochastic search with O(1) expected time for a nearly optimal arm at covariate values in a given set before applying ε-greedy randomization and arm elimination. The procedure is shown to attain the asymptotically minimal rates for the regret over the given set. This chapter is based on Kim et al. (2021) and Lai et al. (2022).The second part is in the area of Markov Chain Monte Carlo (MCMC), and we discuss a new adaptive MCMC algorithm where acceptance rates are significantly improved. The basic idea is to approximate a target distribution by the empirical distribution of V representative atoms, chosen sequentially by an MCMC scheme so that the distribution converges weakly to the target distribution as the number of iterations goes to infinity. Making use of coupling arguments and bounds on the total variation norm of the difference between the target distribution and the empirical measure defined by the sample paths of the MCMC scheme, we establish the asymptotic normality of the Monte Carlo estimate of a functional of the target distribution and provide a consistent estimator of its standard error. This chapter is based on in Lai et al. (2021b).The third part is in the area of object detection, and we build a data-driven methodology for the performance reliability and the improvement of sensor algorithms for automated driving perception tasks. The methodology takes as input three elements: one or various algorithms for object detection when the input is an image, a dataset of camera images that represents a sample from an environment, and a simple policy that serves as a proxy for a task such as driving assistance. We develop a statistical estimator, which combines these elements and a data augmentation technique, in order to rank the reliability of perception algorithms. Reliability is measured as the chance of collision given the speed of the ego vehicle and the distance to the closest object in range. We are able to compare algorithms in the (speed vs distance-to-closest-object) space using p-values and use this information to suggest improved-safety algorithms. This chapter is based on Xu et al. (2021).
- 일반주제명
- Kinematics
- 일반주제명
- Sample size
- 일반주제명
- Dynamic programming
- 일반주제명
- Gaming machines
- 일반주제명
- Cognitive models
- 일반주제명
- Markov analysis
- 일반주제명
- Parameter estimation
- 일반주제명
- Applied mathematics
- 일반주제명
- Computer science
- 일반주제명
- Computational physics
- 키워드
- Kinematics
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2023 us c eng d■001000017358743
■00520260202104745
■006m o d
■007cr#unu||||||||
■020 ▼a9798290652245
■035 ▼a(MiAaPQ)AAI32149746
■035 ▼a(MiAaPQ)Stanfordxh086kx1071
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a500
■1001 ▼aXu, Huanzhong.
■24510▼aMCMC With Substitutions and Multi-Armed Bandits With Covariates: Theory and Applications
■260 ▼a[Sl]▼bStanford University▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a94 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-03, Section: B.
■500 ▼aAdvisor: Mancilla, Jose Blanchet.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2023.
■520 ▼aThe thesis covers three areas of study: Markov Chain Monte Carlo, multi-armed bandits, and reliability of object detection models. Each part begins with a review of relevant literature and a discussion of the motivation behind our novel algorithm. We then present our main theoretical findings, accompanied by proofs detailed in the appendices. Finally, we showcase the results of our simulation studies and real-world applications of our algorithms.The first part is in the area of multi-armed bandits, and we introduce a new approach to nonparametric multi-armed bandit theory involving both the bandit and the covariate processes. The extension to bandit processes with a non-denumerable set of arms is also discussed. The approach we develop herein can be readily extended to continuous-time processes by using ε-greedy randomization and arm elimination instead of dynamic allocation indices. It also carries out a stochastic search with O(1) expected time for a nearly optimal arm at covariate values in a given set before applying ε-greedy randomization and arm elimination. The procedure is shown to attain the asymptotically minimal rates for the regret over the given set. This chapter is based on Kim et al. (2021) and Lai et al. (2022).The second part is in the area of Markov Chain Monte Carlo (MCMC), and we discuss a new adaptive MCMC algorithm where acceptance rates are significantly improved. The basic idea is to approximate a target distribution by the empirical distribution of V representative atoms, chosen sequentially by an MCMC scheme so that the distribution converges weakly to the target distribution as the number of iterations goes to infinity. Making use of coupling arguments and bounds on the total variation norm of the difference between the target distribution and the empirical measure defined by the sample paths of the MCMC scheme, we establish the asymptotic normality of the Monte Carlo estimate of a functional of the target distribution and provide a consistent estimator of its standard error. This chapter is based on in Lai et al. (2021b).The third part is in the area of object detection, and we build a data-driven methodology for the performance reliability and the improvement of sensor algorithms for automated driving perception tasks. The methodology takes as input three elements: one or various algorithms for object detection when the input is an image, a dataset of camera images that represents a sample from an environment, and a simple policy that serves as a proxy for a task such as driving assistance. We develop a statistical estimator, which combines these elements and a data augmentation technique, in order to rank the reliability of perception algorithms. Reliability is measured as the chance of collision given the speed of the ego vehicle and the distance to the closest object in range. We are able to compare algorithms in the (speed vs distance-to-closest-object) space using p-values and use this information to suggest improved-safety algorithms. This chapter is based on Xu et al. (2021).
■590 ▼aSchool code: 0212.
■650 4▼aKinematics
■650 4▼aSample size
■650 4▼aDynamic programming
■650 4▼aGaming machines
■650 4▼aCognitive models
■650 4▼aAtoms & subatomic particles
■650 4▼aMarkov analysis
■650 4▼aParameter estimation
■650 4▼aApplied mathematics
■650 4▼aComputer science
■650 4▼aComputational physics
■653 ▼aKinematics
■653 ▼aDynamic programming
■690 ▼a0364
■690 ▼a0984
■690 ▼a0216
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-03B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358743▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


