國立虎尾科技大學 |

A Novel Approach To Optimization of Iterative Machine Learning Algorithms : = Over Heap Structure.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	A Novel Approach To Optimization of Iterative Machine Learning Algorithms :/
其他題名:	Over Heap Structure.
作者:	Kurban, Hasan.
面頁冊數:	1 online resource (134 pages)
附註:	Source: Dissertation Abstracts International, Volume: 79-04(E), Section: B.
Contained By:	Dissertation Abstracts International79-04B(E).
標題:	Computer science. -
電子資源:	click for full text (PQDT)
ISBN:	9780355342451

A Novel Approach To Optimization of Iterative Machine Learning Algorithms : = Over Heap Structure.
Kurban, Hasan.

A Novel Approach To Optimization of Iterative Machine Learning Algorithms :Over Heap Structure. - 1 online resource (134 pages)

Source: Dissertation Abstracts International, Volume: 79-04(E), Section: B.

Thesis (Ph.D.)

Includes bibliographical references

This thesis describes an optimization approach designed to reduce training run-time complexity of iterative data mining and machine learning algorithms (IT-DMA). As big data becomes truly big, the standard repertoire of data mining and machine learning algorithms over the last several decades have remained virtually unchanged. Despite their age, IT-DMA, such as k-means clustering (KM), expectation maximization for clustering algorithms (EM-T), are still among the most popular learning algorithms and widely used over a variety of domains. However, they become overwhelmed with big data since all data points are being continually and indiscriminately revisited.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2018

Mode of access: World Wide Web

ISBN: 9780355342451Subjects--Topical Terms:

573171
Computer science.
Index Terms--Genre/Form:

554714
Electronic books.

A Novel Approach To Optimization of Iterative Machine Learning Algorithms : = Over Heap Structure.
LDR:03864ntm a2200361Ki 4500 001 910838
005 20180517112612.5
006 m o u
007 cr mn||||a|a||
008 190606s2017 xx obm 000 0 eng d
020 $a 9780355342451
035 $a (MiAaPQ)AAI10635424
035 $a (MiAaPQ)indiana:14945
035 $a AAI10635424
040 $a MiAaPQ $b eng $c MiAaPQ
099 $a TUL $f hyy $c available through World Wide Web
100 1 $a Kurban, Hasan. $3 1182319
245 1 2 $a A Novel Approach To Optimization of Iterative Machine Learning Algorithms : $b Over Heap Structure.
264 0 $c 2017
300 $a 1 online resource (134 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: Dissertation Abstracts International, Volume: 79-04(E), Section: B.
500 $a Adviser: Mehmet M. Dalkilic.
502 $a Thesis (Ph.D.) $c Indiana University $d 2017.
504 $a Includes bibliographical references
520 $a This thesis describes an optimization approach designed to reduce training run-time complexity of iterative data mining and machine learning algorithms (IT-DMA). As big data becomes truly big, the standard repertoire of data mining and machine learning algorithms over the last several decades have remained virtually unchanged. Despite their age, IT-DMA, such as k-means clustering (KM), expectation maximization for clustering algorithms (EM-T), are still among the most popular learning algorithms and widely used over a variety of domains. However, they become overwhelmed with big data since all data points are being continually and indiscriminately revisited.
520 $a In this new era of big data, plentiful memory, and powerful CPUs, IT-DMA eventually are overcome with scale. One is struck by the fact that, as an optimization problem, the data is treated equivocally at each iteration, i.e., no matter the effect on cost: how the data is used remains unchanged and uniform. As data re-visited, however, it is clear that some data has more of a change on cost (high expression) than other (low expression). If there were both a means of assessing this difference as well as their relationship, e.g., does high expression (HE) tend to become low expression (LE), then it could be exploited by guiding the iterate to HE. One especially interesting questions arises: is it possible, or even feasible, to rethink convergence, not as a limit of cost only, but as proportion of HE and LE changing cost? If, for example, the data are all LE, then there is unlikely any substantial change (improvement) to cost.
520 $a In this novel work, we have found a means of answering these questions: we add structure to IT-DMA, separating LE from HE through use of heaps that we call strong and weak. Strong heap possess an additional invariant to the heap property that grows monotonically with insertions. Convergence is found by examining the relative mixes of LE and HE--when no more progress can be made--the leaves remain the same kind, we stop. We show implementation of this framework over two popular IT-DMA algorithms, EM-T and KM. Our results are dramatic improvements over EM-T and KM through different kinds of testing: scale, dimension, and separability. What is as exciting is the question of whether iterative algorithms, like KM, EM, can, in general, be optimized using structures. An interesting side result is that we believe what remains in leaves at convergence is a mix of useful data and noise--data that does not contribute meaningfully to cost.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2018
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 573171
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Indiana University. $b Computer Sciences. $3 1179305
773 0 $t Dissertation Abstracts International $g 79-04B(E).
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10635424 $z click for full text (PQDT)