東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

On Designing Resource-Constrained CNNs Efficiently.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	On Designing Resource-Constrained CNNs Efficiently./
作者:	Chin, Ting-Wu.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	135 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-02, Section: B.
Contained By:	Dissertations Abstracts International83-02B.
標題:	Artificial intelligence. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28646296
ISBN:	9798534664591

On Designing Resource-Constrained CNNs Efficiently.
Chin, Ting-Wu.

On Designing Resource-Constrained CNNs Efficiently. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 135 p.

Source: Dissertations Abstracts International, Volume: 83-02, Section: B.

Thesis (Ph.D.)--Carnegie Mellon University, 2021.

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

Deep Convolutional Neural Networks (CNNs) have been adopted in many computer vision applications to achieve high performance. However, the growing computational demand of CNNs has made it increasingly difficult to deploy state-of-the-art CNNs onto resource-constrained platforms. As a result, model compression/acceleration has emerged to be an important field of research. In this thesis, we intend to make CNNs more friendly for resource-limited platforms from two perspectives. The first perspective is to introduce novel ways of compressing/accelerating CNNs and the second perspective is to reduce the overhead of existing methodologies for constructing resource-constrained CNNs.In the first perspective, we propose one novel technique for model acceleration and another for model compression. First, we propose AdaScale which is an algorithm that automatically scales the resolution of input images to improve both the speed and accuracy of a video object detection system. Second, we identify the Winning-Bitwidth phenomenon, where we found some weight bitwidth is more efficient than others for model compression when the filter counts of the CNNs are allowed to change.In the second perspective, we propose three novel algorithms for accelerating existing filter pruning methods for constructing resource-constrained CNNs. First, we propose LeGR, an algorithm that aims to learn a global ranking among filters of a pre-trained CNN so that compressing the CNN to different target constraint levels using filter pruning can be done efficiently by greedily pruning the filters following the learned ranking. Second, we improve upon LeGR and propose Joslim, which is an algorithm that trains a CNN from scratch by jointly optimizing its weights and filter counts such that the trained CNN can be pruned without fine-tuning. Joslim improves upon LeGR in terms of efficiency as LeGR requires the pruned models to be fine-tuned to be usable. Lastly, we propose Width Transfer, which improves the efficiency for filter pruning methods that are derived from a neural architecture search perspective. Width Transfer assumes that the optimized filter counts are regular across depths and widths of a CNN architecture and are invariant to the size and the resolution of the training dataset. As a result, Width Transfer performs neural architecture search for filter counts by solving a proxy problem that has a much lower overhead.

ISBN: 9798534664591Subjects--Topical Terms:

516317
Artificial intelligence.
Subjects--Index Terms:

Automated machine learning

On Designing Resource-Constrained CNNs Efficiently.
LDR:03627nmm a2200373 4500 001 2344653
005 20220531064617.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798534664591
035 $a (MiAaPQ)AAI28646296
035 $a AAI28646296
040 $a MiAaPQ $c MiAaPQ
100 1 $a Chin, Ting-Wu. $3 3683445
245 1 0 $a On Designing Resource-Constrained CNNs Efficiently.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 135 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-02, Section: B.
500 $a Advisor: Marculescu, Diana;Joshi, Gauri.
502 $a Thesis (Ph.D.)--Carnegie Mellon University, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Deep Convolutional Neural Networks (CNNs) have been adopted in many computer vision applications to achieve high performance. However, the growing computational demand of CNNs has made it increasingly difficult to deploy state-of-the-art CNNs onto resource-constrained platforms. As a result, model compression/acceleration has emerged to be an important field of research. In this thesis, we intend to make CNNs more friendly for resource-limited platforms from two perspectives. The first perspective is to introduce novel ways of compressing/accelerating CNNs and the second perspective is to reduce the overhead of existing methodologies for constructing resource-constrained CNNs.In the first perspective, we propose one novel technique for model acceleration and another for model compression. First, we propose AdaScale which is an algorithm that automatically scales the resolution of input images to improve both the speed and accuracy of a video object detection system. Second, we identify the Winning-Bitwidth phenomenon, where we found some weight bitwidth is more efficient than others for model compression when the filter counts of the CNNs are allowed to change.In the second perspective, we propose three novel algorithms for accelerating existing filter pruning methods for constructing resource-constrained CNNs. First, we propose LeGR, an algorithm that aims to learn a global ranking among filters of a pre-trained CNN so that compressing the CNN to different target constraint levels using filter pruning can be done efficiently by greedily pruning the filters following the learned ranking. Second, we improve upon LeGR and propose Joslim, which is an algorithm that trains a CNN from scratch by jointly optimizing its weights and filter counts such that the trained CNN can be pruned without fine-tuning. Joslim improves upon LeGR in terms of efficiency as LeGR requires the pruned models to be fine-tuned to be usable. Lastly, we propose Width Transfer, which improves the efficiency for filter pruning methods that are derived from a neural architecture search perspective. Width Transfer assumes that the optimized filter counts are regular across depths and widths of a CNN architecture and are invariant to the size and the resolution of the training dataset. As a result, Width Transfer performs neural architecture search for filter counts by solving a proxy problem that has a much lower overhead.
590 $a School code: 0041.
650 4 $a Artificial intelligence. $3 516317
650 4 $a Computer engineering. $3 621879
650 4 $a Computer science. $3 523869
650 4 $a Information technology. $3 532993
650 4 $a Standard deviation. $3 3560390
650 4 $a Accuracy. $3 3559958
650 4 $a Datasets. $3 3541416
650 4 $a Experiments. $3 525909
650 4 $a Optimization. $3 891104
650 4 $a Neural networks. $3 677449
650 4 $a Carbon footprint. $3 3564744
650 4 $a Algorithms. $3 536374
653 $a Automated machine learning
653 $a Convolutional neural networks
653 $a Machine learning for edge devices
653 $a Compression and acceleration
690 $a 0800
690 $a 0464
690 $a 0984
690 $a 0489
710 2 $a Carnegie Mellon University. $b Electrical and Computer Engineering. $3 2094139
773 0 $t Dissertations Abstracts International $g 83-02B.
790 $a 0041
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28646296