Publications

Detailed Information

MM algorithm for sparse regression model with the fused lasso penalty and its parallelization using GPU

DC Field Value Language
dc.contributor.advisor임요한-
dc.contributor.author유동현-
dc.date.accessioned2017-07-14T00:30:46Z-
dc.date.available2017-07-14T00:30:46Z-
dc.date.issued2013-08-
dc.identifier.other000000012429-
dc.identifier.urihttps://hdl.handle.net/10371/121141-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 통계학과, 2013. 8. 임요한.-
dc.description.abstract본 논문에서는 고차원의 fuse lasso 벌점을 지닌 회귀 모형에 대해
GPU를 이용하여 병렬화된 MM 알고리즘을 제안하였다.
본 논문에서 다루는 모형에 대한 MM 알고리즘의 수렴성이 보장됨을 보이고 다양한 형태의 설명 변수의 행렬과 회귀 계수의 구조에 대해서 유연성과 안정성을 가짐을 수치적으로 확인하였다. MM 알고리즘과 기존의 다른 방법들을 다양한 예제들로 비교하여 병렬화된 MM 알고리즘이 차원이 큰 표준의 fused lasso 벌점을 갖는 회귀 모형에서 더 빠르게 추정량을 제공함을 보였다.
-
dc.description.abstractIn this paper, we propose a majorization-minimization (MM) algorithm for high-dimensional fused lasso regression (FLR) suitable for parallelization using graphics processing units (GPUs). The MM algorithm is stable and
exible as it can solve the FLR problems with various types of design matrices and penalty structures within a few tens of iterations. We also show that the convergence of the proposed algorithm is guaranteed. We conduct numerical studies to compare our algorithm with other existing algorithms, demonstrating that the proposed MM algorithm is competitive in many settings including the two-dimensional FLR with arbitrary design matrices. The merit of GPU parallelization is also exhibited.
-
dc.description.tableofcontents1 Introduction 1
2 Review of existing algorithms 6
2.1. Solution path methods 6
2.1.1. Path-wise optimization algorithm 7
2.1.2. Path algorithm for the FLSA 9
2.1.3. Path algorithm for the generalized lasso 11
2.2. First-order methods 16
2.2.1. Ecient fused lasso algorithm 17
2.2.2. Smoothing proximal gradient method 20
2.2.3. Alternating directions methods 23
2.3. Summary of the reviewed algorithms 29
3 MM algorithm for fused lasso problem 31
3.1. MM algorithm for FLR 31
3.2. Convergence 36
4 Parallelization of the MM algorithm with GPU 43
4.1. Parallelization tools 43
4.2. Parallel tridiagonal solvers 45
4.3. Parallelization of the MM algorithm 52
5 Numerical studies 54
5.1. Standard FLR 56
5.2. Two-dimensional FLR 59
6 Conclusion 69
Bibliography 70
Appendix 76
Abstract (in Korean) 90
-
dc.formatapplication/pdf-
dc.format.extent2154554 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subjectfused lasso regression-
dc.subjectMM algorithm-
dc.subjectparallel computation-
dc.subjectgraphics processing unit-
dc.subject.ddc519-
dc.titleMM algorithm for sparse regression model with the fused lasso penalty and its parallelization using GPU-
dc.typeThesis-
dc.description.degreeDoctor-
dc.citation.pages90-
dc.contributor.affiliation자연과학대학 통계학과-
dc.date.awarded2013-08-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share