Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor | Department of Industrial and Systems Engineering | en_US |
dc.contributor.advisor | Ji, Ping (ISE) | en_US |
dc.contributor.advisor | Ren, Jingzheng (ISE) | en_US |
dc.creator | Chen, Leran | - |
dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13648 | - |
dc.language | English | en_US |
dc.publisher | Hong Kong Polytechnic University | en_US |
dc.rights | All rights reserved | en_US |
dc.title | Design and applications of scenario-based machine learning models | en_US |
dcterms.abstract | Since the onset of the Fourth Industrial Revolution, human civilization has accelerated towards a new era defined by digitization and artificial intelligence. In various domains such as industrial production, healthcare, and finance, data has been extensively utilized and generated. The development of technologies like the Internet of Things (IoT) and mobile internet has driven the mass creation and accumulation of rich data. The rapid advancement of machine learning has unlocked even greater value from data; the synergy of algorithms and data has propelled innovation and transformation across numerous industries. | en_US |
dcterms.abstract | In the broad application of machine learning across various domains, several core challenges and limitations persist. (1) The improvement in accuracy relies on complex algorithms and large datasets, making it challenging to balance efficiency and often overlooking the performance of individual object. (2) The advancement of computing power lags behind the rapid generation of data, limiting the full utilization of available data. (3) Low interpretability and customization hinder user acceptance. These issues restrict the broader application and integration of machine learning technologies. | en_US |
dcterms.abstract | To address the above challenges, this thesis analyzes an observation that models emphasizing generalizability often face difficulties in balancing various performance metrics within specific applications. This research critically examines the conventional emphasis on generalization ability within the existing "Impossible Trinity" of machine learning, i.e. generalizability, efficiency, and accuracy. A framework based on scenario customization is proposed, shifting the focus from generalizability to adaptability and customization while simultaneously pursuing both efficiency and accuracy. The key contributions of this work are as follows: | en_US |
dcterms.abstract | Machine learning models tailored to a single individual object. This approach focuses on the individualized analysis of each object within the scenario to achieve customization for specific scenarios. Subsequent experiments conducted on a Hepatitis C dataset demonstrate that the customized approach achieved an accuracy of 99.49%, surpassing general-purpose algorithms by 5%. | en_US |
dcterms.abstract | Machine learning models based on the specific requirements of the application scenario. This effort primarily aims to achieve optimal results in scenarios by balancing multiple objective functions based on the specific requirements of the scenario. This effort integrates customization for individual people and considers the constraints of actual scenarios. Simulations on the UCI heart disease dataset, mimicking situations of large-scale screening, achieved an accuracy of 97%, proving the efficacy in model lightweight, privacy protection, and model updates with efficient data utilization. | en_US |
dcterms.abstract | Machine learning models inspired by core characteristic of special scenarios. This approach, through a thorough understanding of actual usage scenarios, identifies and fully leverages the core characteristic of scenarios to simplify problems fundamentally and enhance model performance. In the application of obstacle detection in high-speed railway scenarios, images experiments conducted in the Carla simulation environment showed that detection accuracy improved by 10%, and detection speed increased by up to 75%. With point cloud data, the system achieved unparalleled performance with matching accuracy exceeding 96% and speeds of 116 frames per second. | en_US |
dcterms.abstract | In conclusion, this thesis advances improved machine learning performance by addressing model efficiency, accuracy, and adaptability with scenario-specific solutions. These innovations enrich both the theoretical and practical aspects of the field, promising more personalized and efficient future applications. | en_US |
dcterms.extent | xiv, 340 pages : color illustrations | en_US |
dcterms.isPartOf | PolyU Electronic Theses | en_US |
dcterms.issued | 2025 | en_US |
dcterms.educationalLevel | Ph.D. | en_US |
dcterms.educationalLevel | All Doctorate | en_US |
dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13648