國立虎尾科技大學 |

Domain Generalization via Representation Learning.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Domain Generalization via Representation Learning./
作者:	Lyu, Boyang.
面頁冊數:	1 online resource (153 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-08, Section: A.
Contained By:	Dissertations Abstracts International85-08A.
標題:	Information science. -
電子資源:	click for full text (PQDT)
ISBN:	9798381701586

Domain Generalization via Representation Learning.
Lyu, Boyang.

Domain Generalization via Representation Learning. - 1 online resource (153 pages)

Source: Dissertations Abstracts International, Volume: 85-08, Section: A.

Thesis (Ph.D.)--Tufts University, 2024.

Includes bibliographical references

The success of traditional machine learning techniques is highly dependent on the assumption that the training and test data are drawn from independent and identical distributions. This assumption provides a theoretical guarantee and serves as a foundation for the high performance of traditional methods. However, in real-world applications, the performance of well-trained models often suffers from degradation due to the violation of this fundamental assumption. Factors such as environment, equipment, and human activity can easily lead to significant differences in data distribution across training and test data, which poses challenges to the generalization ability of models.One direction of addressing the above problem is Domain Generalization, which aims to enhance the generalization ability of trained models, allowing them to perform well even on unseen test data with different distributions. In this thesis, we conduct a comprehensive review of previous work, focusing on the theoretical foundations, algorithms, and workflows associated with the representation learning-based domain generalization algorithms. We identify the gap between previous theoretical work and practical algorithms, and propose a novel theory to bridge this gap. We also explore the weakness of some domain generalization principles and propose an algorithm as a potential solution. In addition to focusing on algorithms, we recognize the importance of model selection for methods designed for domain generalization. In light of this, we propose a novel model selection method that takes into account the unique characteristics and challenges associated with domain generalization. This selection method considers the complexities of domain shifts and ensures the reliable assessment of model generalization across different domains. By incorporating this validation method into the evaluation process, we can gain more insights into the application of domain generalization algorithms to practical problems.Through this thesis, we contribute to the field of domain generalization by bridging the gap between previous theory and practice, offering potential solutions to address the failure cases observed in certain domain generalization methods and emphasizing the importance of considering the workflow of the domain generalization problem. The proposed theoretical advancements, algorithms, and validation method collectively aim to enable machine learning models to generalize effectively across diverse real-world domains with varying data distributions.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798381701586Subjects--Topical Terms:

561178
Information science.
Subjects--Index Terms:

AlgorithmsIndex Terms--Genre/Form:

554714
Electronic books.

Domain Generalization via Representation Learning.
LDR:03915ntm a22004097 4500 001 1142883
005 20240429063857.5
006 m o d
007 cr mn ---uuuuu
008 250605s2024 xx obm 000 0 eng d
020 $a 9798381701586
035 $a (MiAaPQ)AAI30816719
035 $a AAI30816719
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Lyu, Boyang. $3 1467403
245 1 0 $a Domain Generalization via Representation Learning.
264 0 $c 2024
300 $a 1 online resource (153 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: 85-08, Section: A.
500 $a Advisor: Aeron, Shuchin.
502 $a Thesis (Ph.D.)--Tufts University, 2024.
504 $a Includes bibliographical references
520 $a The success of traditional machine learning techniques is highly dependent on the assumption that the training and test data are drawn from independent and identical distributions. This assumption provides a theoretical guarantee and serves as a foundation for the high performance of traditional methods. However, in real-world applications, the performance of well-trained models often suffers from degradation due to the violation of this fundamental assumption. Factors such as environment, equipment, and human activity can easily lead to significant differences in data distribution across training and test data, which poses challenges to the generalization ability of models.One direction of addressing the above problem is Domain Generalization, which aims to enhance the generalization ability of trained models, allowing them to perform well even on unseen test data with different distributions. In this thesis, we conduct a comprehensive review of previous work, focusing on the theoretical foundations, algorithms, and workflows associated with the representation learning-based domain generalization algorithms. We identify the gap between previous theoretical work and practical algorithms, and propose a novel theory to bridge this gap. We also explore the weakness of some domain generalization principles and propose an algorithm as a potential solution. In addition to focusing on algorithms, we recognize the importance of model selection for methods designed for domain generalization. In light of this, we propose a novel model selection method that takes into account the unique characteristics and challenges associated with domain generalization. This selection method considers the complexities of domain shifts and ensures the reliable assessment of model generalization across different domains. By incorporating this validation method into the evaluation process, we can gain more insights into the application of domain generalization algorithms to practical problems.Through this thesis, we contribute to the field of domain generalization by bridging the gap between previous theory and practice, offering potential solutions to address the failure cases observed in certain domain generalization methods and emphasizing the importance of considering the workflow of the domain generalization problem. The proposed theoretical advancements, algorithms, and validation method collectively aim to enable machine learning models to generalize effectively across diverse real-world domains with varying data distributions.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Information science. $3 561178
650 4 $a Electrical engineering. $3 596380
650 4 $a Computer science. $3 573171
653 $a Algorithms
653 $a Domain generalization
653 $a Machine learning
653 $a Wasserstein distance
653 $a Neural network
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0544
690 $a 0723
690 $a 0800
710 2 $a Tufts University. $b Electrical Engineering. $3 1184138
710 2 $a ProQuest Information and Learning Co. $3 1178819
773 0 $t Dissertations Abstracts International $g 85-08A.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30816719 $z click for full text (PQDT)