| Author: | Lin, Wu |
| Title: | Towards adaptive knowledge transfer in evolutionary transfer optimization |
| Advisors: | Tan, Kay Chen (DSAI) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Subject: | Evolutionary programming (Computer science) Genetic algorithms Mathematical optimization Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Computing |
| Pages: | xvi, 185 pages : color illustrations |
| Language: | English |
| Abstract: | Evolutionary transfer optimization (ETO) is an emerging search paradigm, which integrates evolutionary algorithms (EAs) with transfer learning techniques. Learning and transferring useful knowledge across related problems can reduce repeated searches, enabling traditional EAs to achieve better optimization efficiency and performance on various complex problems. Generally, the design of ETO approaches grapples with three critical issues concerning knowledge transfer: 1) what to transfer, 2) how to transfer, and 3) when to transfer. Considering what to transfer, it involves identifying the type of knowledge and deciding which one to transfer among all available candidates. Regarding how to transfer, it focuses on the methodology design for implementing knowledge transfer. As for the issue of when to transfer, it aims to identify the optimal timing or the appropriate extent for deciding how much knowledge to transfer. However, in existing ETO studies, most deterministic methods lack the adaptability and flexibility when addressing the above three issues, severely limiting the robustness and effectiveness of knowledge transfer in enhancing the optimization efficiency and performance of EAs. To achieve more effective and robust performance, this thesis focuses on studying and designing adaptive knowledge transfer methods to intelligently address one or more of the three issues. Firstly, to adaptively decide what to transfer, this thesis proposes a fuzzy classifier-assisted solution transfer method to identify the most useful solution for transfer in evolutionary sequential transfer optimization (ESTO). By constructing the training data, the fuzzy classifier is built to estimate the solution usefulness of all available source tasks for the target task. Compared to existing solution transfer methods, the proposed method not only estimates whether one source task is useful or useless but also further quantifies the degree of its usefulness when it is estimated to be useful. In this way, the most useful solution is accurately selected from useful source tasks for knowledge transfer. This effectively accelerates the optimization of the target task by adaptively selecting the most useful solution from available source tasks for knowledge transfer. Secondly, to adaptively decide how to transfer, this thesis proposes an ensemble method to combine multiple domain adaptation methods for evolutionary multitasking (EMT), mitigating the unique biases of each domain adaptation method. It smartly addresses the balance between efficacy and diversity when determining which one domain adaptation method for use, which further enhances knowledge transferability across tasks in EMT. The proposed methodology clearly differentiates from existing ensemble methods by integrating a novel adaptive selection mechanism that considers both past performance and diversity of candidate domain adaptation methods. This could potentially lead to more robust and effective multitasking performance in comparison with existing non-adaptive approaches that adopt one deterministic domain adaptation method to address the issue of how to transfer. Lastly, to adaptively decide when to transfer and how to transfer, this thesis proposes a fuzzy logic-based method in EMT. The proposed method includes two fuzzy logic-based components. To effectively adapt the transfer extent along the multitasking search process, a fuzzy logic-based parameter adaption component is developed to dynamically adjust the value of the transfer parameter, thereby alleviating the risk of negative transfer. To adaptively select the most promising method for knowledge transfer, a fuzzy logic-based selection component is developed to select the optimal transfer method from multiple candidates, thereby enhancing knowledge transferability across tasks. The proposed fuzzy logic-based methodology clearly differentiates from existing methods by employing fuzzy logic to effectively process fuzzy and inaccurate information, facilitating effective knowledge transfer. |
| Rights: | All rights reserved |
| Access: | open access |
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/200/13597