상세정보
Export to Refworks부가기능
Robust and Scalable Algorithms for Bayesian Nonparametric Machine Learning
상세 프로파일
| 자료유형 | 학위논문 |
|---|---|
| 서명/저자사항 | Robust and Scalable Algorithms for Bayesian Nonparametric Machine Learning. |
| 개인저자 | Roychowdhury, Anirban. |
| 단체저자명 | The Ohio State University. Computer Science and Engineering. |
| 발행사항 | [S.l.]: The Ohio State University., 2017. |
| 발행사항 | Ann Arbor: ProQuest Dissertations & Theses, 2017. |
| 형태사항 | 188 p. |
| 기본자료 저록 | Dissertation Abstracts International 79-10B(E). Dissertation Abstract International |
| ISBN | 9780438091412 |
| 학위논문주기 | Thesis (Ph.D.)--The Ohio State University, 2017. |
| 일반주기 |
Source: Dissertation Abstracts International, Volume: 79-10(E), Section: B.
Adviser: Srinivasan Parthasarathy. |
| 요약 | Bayesian nonparametric techniques provide a rich set of tools for modeling complex probabilistic machine learning problems. However the richness comes at the cost of significant complexity of learning and inference for large scale datasets, in a |
| 요약 | First, we develop fast inference algorithms for sequential models with Bayesian nonparametric priors using small-variance asymptotics, an emerging technique for obtaining scalable combinatorial algorithms from rich probabilistic models. We deriv |
| 요약 | We start the second section with a novel stick-breaking definition of a certain class of Bayesian nonparametric priors called gamma processes (GP), using its characterization as a completely random measure and attendant Poisson process machinery |
| 요약 | In the third section, we use concepts from statistical physics to develop a robust Monte Carlo sampler that efficiently traverses the parameter space. Built on the Hamiltonian Monte Carlo framework, our sampler uses a modified Nose-Poincare Hami |
| 요약 | We continue with an L-BFGS optimization algorithm on Riemannian manifolds that uses stochastic variance reduction techniques for fast convergence with constant step sizes, without resorting to standard linesearch methods, and provide a new conve |
| 요약 | We finish with a novel technique for learning the mass matrices in Monte Carlo samplers obtained from discretized dynamics that preserve some energy function, by using existing dynamics in the sampling step of a Monte Carlo EM framework, and lea |
| 일반주제명 | Computer science. |
| 언어 | 영어 |
| 바로가기 | 
: 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
