國立高雄大學圖資館 |

Language: English

Back

Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks /Sayeed Shafayet Chowdhury.
Author:	Chowdhury, Sayeed Shafayet,
Description:	1 electronic resource (194 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
Notes:	Advisors: Roy, Kaushik Committee members: Raghunathan, Anand; Sen, Shreyas; Raghunathan, Vijay.
Contained By:	Dissertations Abstracts International86-05B.
Subject:	Syntax.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31682050
ISBN:	9798346502135

Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks /
Chowdhury, Sayeed Shafayet,

Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks /Sayeed Shafayet Chowdhury. - 1 electronic resource (194 pages)

Source: Dissertations Abstracts International, Volume: 86-05, Section: B.

Today's deep neural networks (DNNs) have achieved tremendous performance in various domains such as computer vision, natural language processing, robotics, generative tasks etc. However, these high-performing DNNs require enormous amounts of compute, resulting in significant power consumption. Moreover, these often struggle in terms of visual understanding capabilities. To that effect, this thesis focuses on two aspects - enhancing efficiency of neural networks and improving their visual understanding. On the efficiency front, we leverage brain-inspired Spiking Neural Networks (SNNs), which offer a promising alternative to traditional deep learning. We first perform a comparative analysis between models with and without leak, revealing that leaky-integrate-and-fire (LIF) model provides improved robustness and better generalization compared to integrate-and-fire (IF). However, leak decreases the sparsity of computation. In the second work, by introducing a Discrete Cosine Transform-based novel spike encoding scheme (DCT-SNN), we demonstrate significant performance improvements, achieving 2-14X reduction in latency compared to state-of-the-art SNNs. Next, a novel temporal pruning method is proposed, which dynamically reduces the number of timesteps during training, enabling SNN inference with just one timestep while maintaining high accuracy. The second focus of the thesis is on improving the visual understanding aspect of DNNs. The first work along this direction introduces a framework for visual syntactic understanding, drawing parallels between linguistic syntax and visual components of an image. By manipulating images to create syntactically incorrect examples and using a BERT-like autoencoder for reconstruction, the study significantly enhances the visual syntactic recognition capabilities of DNNs, evidenced by substantial improvements in classification accuracies on the Celeb: A and AFHQ datasets. Further, the thesis tackles unsupervised procedure learning from videos, given multiple videos of the same underlying task. Employing optimal transport (OT) and introducing novel regularization strategies, we develop the 'OPEL' framework, which substantially outperforms existing methods (27-46% average enhancement in F1-score) on both egocentric and third-person benchmarks. Overall, the dissertation advances the field by proposing brain-inspired models and novel learning frameworks that significantly enhance the efficiency and visual understanding capabilities of deep learning systems, making them more suitable for real-world applications.

English

ISBN: 9798346502135Subjects--Topical Terms:

275646
Syntax.

Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks /
LDR:03941nmm a22003853i 4500 001 676406
005 20250420214400.5
006 m o d
007 cr|nu||||||||
008 250901s2024 miu||||||m |||||||eng d
020 $a 9798346502135
035 $a (MiAaPQD)AAI31682050
035 $a (MiAaPQD)Purdue26811904
035 $a AAI31682050
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Chowdhury, Sayeed Shafayet, $e author. $3 990236
245 1 0 $a Enhancing Visual Understanding and Energy-Efficiency in Deep Neural Networks / $c Sayeed Shafayet Chowdhury.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (194 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: Dissertations Abstracts International, Volume: 86-05, Section: B.
500 $a Advisors: Roy, Kaushik Committee members: Raghunathan, Anand; Sen, Shreyas; Raghunathan, Vijay.
502 $b Ph.D. $c Purdue University $d 2024.
520 $a Today's deep neural networks (DNNs) have achieved tremendous performance in various domains such as computer vision, natural language processing, robotics, generative tasks etc. However, these high-performing DNNs require enormous amounts of compute, resulting in significant power consumption. Moreover, these often struggle in terms of visual understanding capabilities. To that effect, this thesis focuses on two aspects - enhancing efficiency of neural networks and improving their visual understanding. On the efficiency front, we leverage brain-inspired Spiking Neural Networks (SNNs), which offer a promising alternative to traditional deep learning. We first perform a comparative analysis between models with and without leak, revealing that leaky-integrate-and-fire (LIF) model provides improved robustness and better generalization compared to integrate-and-fire (IF). However, leak decreases the sparsity of computation. In the second work, by introducing a Discrete Cosine Transform-based novel spike encoding scheme (DCT-SNN), we demonstrate significant performance improvements, achieving 2-14X reduction in latency compared to state-of-the-art SNNs. Next, a novel temporal pruning method is proposed, which dynamically reduces the number of timesteps during training, enabling SNN inference with just one timestep while maintaining high accuracy. The second focus of the thesis is on improving the visual understanding aspect of DNNs. The first work along this direction introduces a framework for visual syntactic understanding, drawing parallels between linguistic syntax and visual components of an image. By manipulating images to create syntactically incorrect examples and using a BERT-like autoencoder for reconstruction, the study significantly enhances the visual syntactic recognition capabilities of DNNs, evidenced by substantial improvements in classification accuracies on the Celeb: A and AFHQ datasets. Further, the thesis tackles unsupervised procedure learning from videos, given multiple videos of the same underlying task. Employing optimal transport (OT) and introducing novel regularization strategies, we develop the 'OPEL' framework, which substantially outperforms existing methods (27-46% average enhancement in F1-score) on both egocentric and third-person benchmarks. Overall, the dissertation advances the field by proposing brain-inspired models and novel learning frameworks that significantly enhance the efficiency and visual understanding capabilities of deep learning systems, making them more suitable for real-world applications.
546 $a English
590 $a School code: 0183
650 4 $a Syntax. $3 275646
650 4 $a Spectrum analysis. $3 188665
650 4 $a Neural networks. $3 915594
650 4 $a Analytical chemistry. $3 708695
650 4 $a Optics. $3 204142
690 $a 0486
690 $a 0800
690 $a 0752
710 2 $a Purdue University. $3 966863
720 1 $a Roy, Kaushik $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-05B.
790 $a 0183
791 $a Ph.D.
792 $a 2024
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31682050