國立高雄大學圖資館 |

Language: English

Back

Statistics, Computation, and Adaptat...

Pananjady Martin, Ashwin.

Statistics, Computation, and Adaptation in High Dimensions.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Statistics, Computation, and Adaptation in High Dimensions.
Author:	Pananjady Martin, Ashwin.
Published:	Ann Arbor : ProQuest Dissertations & Theses, 2020
Description:	339 p.
Notes:	Source: Dissertations Abstracts International, Volume: 82-06, Section: B.
Notes:	Advisor: Wainwright, Martin J.;Courtade, Thomas A.
Contained By:	Dissertations Abstracts International82-06B.
Subject:	Statistics.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28090238
ISBN:	9798691238284

Statistics, Computation, and Adaptation in High Dimensions.
Pananjady Martin, Ashwin.

Statistics, Computation, and Adaptation in High Dimensions. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 339 p.

Source: Dissertations Abstracts International, Volume: 82-06, Section: B.

Thesis (Ph.D.)--University of California, Berkeley, 2020.

This item must not be sold to any third party vendors.

With a focus on designing flexible, tractable, and adaptive methodology for some canonical machine learning tasks, we establish several results for the class of permutation-based models, index models, and Markov reward processes. First, we study permutation-based models in the vector, matrix, and tensor settings, which provide robust representations in "broken-sample" problems and of human-generated data. We design tractable and adaptive methodological solutions for fitting these models that, among other things, narrow statistical computational gaps conjectured in the literature. Second, we study a subclass of index models–widely used in dimensionality reduction and exploratory data analysis–through a computational lens, focusing on avoiding the (statistical) curse of dimensionality and on achieving automatic adaptation to the noise level in the problem. Our perspective yields efficient algorithms for solving these non-convex fitting problems that come with provable guarantees of sample efficiency and adaptation. Finally, we turn to studying some statistical questions in reinforcement learning, focusing in particular on instance-dependent guarantees for the policy evaluation problem. We show that while some algorithms attain the optimal, "local" performance for this problem, other popular methods fall short and must be modified in order to achieve the desired levels of adaptation.

ISBN: 9798691238284Subjects--Topical Terms:

182057
Statistics.
Subjects--Index Terms:

Adaptive estimation

Statistics, Computation, and Adaptation in High Dimensions.
LDR:02622nmm a2200373 4500 001 594615
005 20210521101708.5
008 210917s2020 ||||||||||||||||| ||eng d
020 $a 9798691238284
035 $a (MiAaPQ)AAI28090238
035 $a AAI28090238
040 $a MiAaPQ $c MiAaPQ
100 1 $a Pananjady Martin, Ashwin. $3 886657
245 1 0 $a Statistics, Computation, and Adaptation in High Dimensions.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 339 p.
500 $a Source: Dissertations Abstracts International, Volume: 82-06, Section: B.
500 $a Advisor: Wainwright, Martin J.;Courtade, Thomas A.
502 $a Thesis (Ph.D.)--University of California, Berkeley, 2020.
506 $a This item must not be sold to any third party vendors.
520 $a With a focus on designing flexible, tractable, and adaptive methodology for some canonical machine learning tasks, we establish several results for the class of permutation-based models, index models, and Markov reward processes. First, we study permutation-based models in the vector, matrix, and tensor settings, which provide robust representations in "broken-sample" problems and of human-generated data. We design tractable and adaptive methodological solutions for fitting these models that, among other things, narrow statistical computational gaps conjectured in the literature. Second, we study a subclass of index models–widely used in dimensionality reduction and exploratory data analysis–through a computational lens, focusing on avoiding the (statistical) curse of dimensionality and on achieving automatic adaptation to the noise level in the problem. Our perspective yields efficient algorithms for solving these non-convex fitting problems that come with provable guarantees of sample efficiency and adaptation. Finally, we turn to studying some statistical questions in reinforcement learning, focusing in particular on instance-dependent guarantees for the policy evaluation problem. We show that while some algorithms attain the optimal, "local" performance for this problem, other popular methods fall short and must be modified in order to achieve the desired levels of adaptation.
590 $a School code: 0028.
650 4 $a Statistics. $3 182057
650 4 $a Computer science. $3 199325
650 4 $a Electrical engineering. $3 454503
653 $a Adaptive estimation
653 $a Index models
653 $a Permutation-based models
653 $a Reinforcement learning
653 $a Statistical-computational tradeoffs
690 $a 0463
690 $a 0984
690 $a 0544
710 2 $a University of California, Berkeley. $b Electrical Engineering & Computer Sciences. $3 886608
773 0 $t Dissertations Abstracts International $g 82-06B.
790 $a 0028
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28090238