博士学位論文

(1)

博

士

学

位

論

文

Doctoral Dissertation

内容の要旨及び審査結果の要旨

Dissertation Abstract and

Summary of the Dissertation Review Result

第

31

号

The Thirty-First Issue

平成

30

年

3

月

March, 2018

The University of Aizu

(2)

はしがき

博士の学位を授与したので、学位規則（昭和２８年４月１日文部省令第９号）第８条の規定に基づき、その論文の内容の要旨及び論文審査の結果の要旨をここに公表する。

学位記番号に付した｢甲｣は学位規則第４条第１項（いわゆる課程博士）によるものであることを示す。また、「乙」は学位規則第４条第２項（いわゆる論文博士）によるものであることを示す。

Preface

On granting the Doctoral Degree to the individuals mentioned below, abstracts of their theses and the theses review results are herewith publicly announced, in according to the provisions provided for in Article 8 of the Ruling of Degrees (Ministry Of Education Ordinance No.9, enacted on April 1, 1953)

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase” or the Doctoral

Degree granted by the University at which the grantee was enrolled.). The Chinese

character,

“乙”, at the beginning of the diploma number represents that an individual has

been granted the degree in accordance with the provisions provided for in Paragraph 4 -2 of

the Ruling Of Degrees (what is called

“Ronpaku”).

(3)

- 1 -

Associate

Prof. YEN, N. Y.

会津大学上級准教授モハメドハマダ（主査）

会津大学教授丁数学

会津大学教授ヴィタリークリュエフ会津大学准教授嚴昱文

(5)

- 3 -

Abstract

Recommender systems are intelligent decision support software tools that help users to discover items that might be of interest to them. They assist users to make decisions on a variety of items from different sources. Various techniques and approaches have been applied to design and implement such systems to generate credible recommendations to users. Traditionally, the most common techniques used by many existing recommendation systems are collaborative filtering, content-based filtering, knowledge-based filtering, and hybrid-based filtering that combines two or more techniques in different ways. A multi-criteria recommendation is an extended technique used to predict unknown ratings and recommend exciting items to users based on ratings given to multiple attributes of items.

This method has been used and proven by researchers in industries and academic institutions to provide more accurate predictions than the traditional techniques.

Accuracy improvement has been one of the issues yet to be solved by recommender systems research community. Recently, multi-criteria recommender systems that use multiple criteria ratings to estimate overall rating have been receiving considerable attention within the recommender systems research domain. What is still not yet clear is the role of some sophisticated machine learning and optimization algorithms such as artificial neural networks, genetic algorithms, simulated annealing algorithms, and other algorithms that could improve the accuracy of the system. The current study proposes new machine learning methods for modeling preferences of users based on several attributes of the items.

The first proposed method is to develop a methodological framework based on artificial neural networks to model preferences of users by integrating the neural network with collaborative filtering techniques for predicting unknown ratings and providing recommendations of the most preferred items for users. The neural network was trained using simulated annealing algorithms and integrated with a single-rating recommender system.

The second proposed framework is based on genetic algorithms for predicting user preferences in multi-criteria recommendation systems. In this framework, three genetic algorithms, namely the standard genetic algorithm, the adaptive genetic algorithm, and the multi-heuristic genetic algorithm are adopted to conduct the experiments. The three genetic algorithms are used separately to understand the predictive performance of our proposed framework.

To demonstrate the effectiveness of our proposed frameworks, several experiments were carried out.

In the experiments, our proposed model was demonstrated as movie and hotel recommender systems and their performance was tested with real user data for movies and hotel recommendations. The real user data for movies and hotels were extracted from Yahoo!Movie and TripAdvisor respectively, and we used them to simulate the interactions between real users and the systems.

The experimental results for each of the two proposed frameworks together with their corresponding single rating techniques are presented in this study. To analyze the performance of the approaches, we carried out a comparative analysis of their performance with the collaborative filtering technique and other multi-criteria recommendation methods. Through a comparison study with other single- and multi-criteria collaborative filtering methodologies, we demonstrated that using the proposed machine learning method is an integral part of the multi-criteria recommendation process. In addition, the

(6)

- 4 -

experimental results also show that using artificial neural networks to incorporate the multi-criteria rating information for predicting the overall rating has considerably improved the accuracy of the multi-criteria recommender system than using genetic algorithms.

Furthermore, we explored other neural network models by increasing the size of the neural networks and using gradient descent-based techniques to train multi-layered networks. To do this, we proposed to simulate the performance of the multi-layered network trained with backpropagation and Levenberg Marquardt algorithms, and a single-layer network trained with delta rule and compare their prediction accuracy. The operational results of the experiments for training and testing these neural networks using the Yahoo!Movie dataset are also presented. The results have shown that the neural network trained with delta rule has faster convergence and higher prediction accuracy than the multi-layered neural network trained with the backpropagation and Levenberg Marquardt algorithms.

Summary of the Dissertation Review Result

In this dissertation, the candidate proposed new machine learning methods for modeling preferences of users in a multi-criteria recommender system. He proposed two methods of incorporating the multiple criteria rating information for estimating the overall preferences of users of the system.

In the first proposed method, he developed a methodological framework based on artificial neural networks to model preferences of users by integrating the neural network with collaborative filtering techniques for predicting unknown ratings and providing recommendations of the most preferred items for users. The neural network was trained using simulated annealing algorithms and integrated with a single-rating recommender system.

The second method was proposed based on genetic algorithms for predicting user preferences in multi-criteria recommendation systems. In this framework, the candidate adopted three genetic algorithms, namely the standard genetic algorithm, the adaptive genetic algorithm, and the multi-heuristic genetic algorithm to conduct the experiments. The three genetic algorithms are used separately to understand the predictive performance of our proposed framework.

To demonstrate the effectiveness of the proposed methods, several experiments have been carried out. In the experiments, the proposed methods were demonstrated as movie and hotel recommender systems, and their performance was tested with real-user data for movies and hotel recommendations.

The real-user data for movies and hotels were extracted from Yahoo!Movie and TripAdvisor websites respectively.

The dissertation and presentation were well organized, and the candidate presented his work and the results obtained. He published four major-journal papers, three major conference papers, and two non-major conference papers. All the publications are related to the dissertation. He has sufficient English ability and communicates with the referees fluently. After questions and answers session, the candidate was asked to go out, and committee members unanimously agreed that the candidate passes the final review of his doctoral thesis.

(7)

- 5 - Name

氏名

閆宇

YAN, Yu （イエンユー）

甲CI博第63号

論文題目

Programming Learning Support Methods based on Adaptive Hypermedia and Program Visualization

適応型ハイパーメディアとプログラム可視化を用いたプログラミング学習支援手法の研究

Senior Associate Prof. KUROKAWA, H. (Chief Referee)

Senior Associate Prof. YOSHIOKA, R.

The University of Aizu, Associate Prof. JING, L.

The University of Aizu, Associate Prof. PEI, Y.

会津大学上級准教授黒川弘国（主査）

会津大学上級准教授吉岡廉太郎会津大学准教授荊雷

会津大学准教授裴岩

(8)

- 6 -

Abstract

Computer programming skill is required by various fields for the development of information society. To better prepare students for research and work, computer programming has been implemented for not only the field of computer science but also other fields in a growing number of universities on a global scale. In some countries such as the UK and Japan, high school students and even elementary school students are expected to understand basic knowledge about algorithms and simple programs. However, some beginners are having difficulty learning computer programming.

One of the problems is that beginners have difficulties understanding the behavior of a program at runtime and the concepts included in the program. In addition, there is a lack of learning materials that are well constructed and appropriate for each individual beginner. Addressing those issues, research on computer programming education has proposed many educational support methods that utilize new information and communication technology and lead to learning support tools. In particular, methods based on Program Visualization (PV) and Adaptive Hypermedia (AH) technology are popular now.

PV is for static and dynamic graphical representations of computer programs and processed data.

Educational PV tools can graphically show the behaviors of programs at runtime, step by step, to help instructors and students in computer programming education activities. AH provides links or content most appropriate to the current user according to a model of the user's goals, preferences and knowledge. The basis of AH-based support methods is to construct Personalized E-learning for Programming (PEP) to recommend individual beginners the most appropriate learning content or learning plan.

This research focuses on helping beginners to learn computer language syntax and enhancing programming lectures and PEP. It proposed the following educational method to solve the issues in existing PV tools and PEP.

A Lecture-Oriented PV Tool (LOPV tool) is for promoting students' understanding of program syntax and concepts during lectures with the following features: (1) It does not show any information to students that is irrelevant to teaching contents when it visualizes target programs, (2) it is directly launched from teaching contents with the programs to be visualized, (3) it does not require any change in the appearance of the target program source code, (4) it shows visualization results with high enough visibility and more easily distinguishable expressions, (5) it offers enough visualizations so that beginners can learn all of the beginner- level program syntax, and (6) it provides enough functions that are convenient for instructors to operate in the lectures.

Syntactic Knowledge Point based Personalized E-learning for Programming (SKP- based PEP) is an approach to implement the PEP model. Syntactic Knowledge Point (SKP) is a way to describe the syntactic knowledge contained in a program source code. SKP-based PEP monitors individual learners' learning statuses by estimating their understanding of each type of syntactic knowledge. In addition, SKP allows instructors to check the rationality of the knowledge distribution in their learning materials in a more convenient way. Furthermore, it can recommend appropriate learning contents for programming to each learner.

Since 2015, PROVIT (PROgram VIsualization Tool), an experimental LOPV tool developed in the

(9)

- 7 -

author's university, has been consecutively and actively used by the instructors to enhance their lectures. The results of the experimental application in an introductory C programming course show that the LOPV tool is necessary to support lectures on programming. For SKP- based PEP, numerical experiments were conducted and the results of the experiments verify the correctness of the proposed algorithms.

Summary of the Dissertation Review Result

This research gives solutions for the issues in today's computer programming education. This research has two main contributions. The first contribution is a method, "Lecture-oriented program visualization(LOPV)", for helping instructors to promote students' understanding of program language and concepts in lecture. This method has been implemented by an experimental system PROVIT (PRogram VIsualization Tool). PROVIT has been used in a regular introductory C programming course, "Introduction of Programming", in the University of Aizu since 2015. The importance and usefulness of LOPV has been confirmed by the evaluation results of a 2-year experiment using PROVIT. The second contribution is a method for personalized programming learning contents recommendation that the whole process of individual learner status estimation based on his feedback (e.g. example programs he read or ran; answer program of exercise he submitted), and learning contents recommendation can be completely handled by computer system without giving additional burden to the educators. The feasibility of this method has been evaluated by a simulation experiment.

Those two contributions have been published as two major journal papers and four major conference papers. The candidate is the first author of all these papers.

In the final review, the candidate has improved her presentation and dissertation according to the review committee's comments. In conclusion, the candidate has fulfilled all of the formal requirements for the doctoral degree.

(10)

- 8 - Name

氏名

王堃

WANG, Kun （ワンクン）

乙論博第5号

論文題目

Eavesdropping Defense with Jamming for Wireless Networks using Game Theory

ゲーム理論を用いた無線ネットワークのための妨害による盗聴防止

The University of Aizu, Prof. MIYAZAKI, T. (Chief Referee)

The University of Aizu, Prof. TEI, S.

The University of Aizu, Prof. PHAM, A.

The University of Aizu, Senior Associate Prof. TRUONG, C.T.

The Hong Kong Polytechnic University, Prof. GUO, S.

会津大学教授宮崎敏明（主査）

会津大学教授程子学

会津大学教授ファントゥアンアン会津大学上級准教授コンタンチョオン香港理工大学教授ソンゴオ

(11)

- 9 -

Abstract

With the development of communication technology, wireless networks are widely deployed in our lives. The wireless terminals such as mobile phone and laptop will lead to complicated traffic environment. These wireless terminals are allowed to join and leave the system freely. Note that we cannot guarantee that they are friendly and the aim of them may be to destroy the information transmission process. Therefore, the wireless networks are highly vulnerable and susceptible to eavesdropping attacks. Anyone who participates in the communication can listen and possibly extract information.

Based on this, we pay attention to the physical layer security, which is proposed as a complementary method of higher-layer techniques. In this dissertation, we study the eavesdropping defense problem in the presence of intermediate nodes. The intermediate nodes can act as jammer or relay. The jammer is used to broadcast artificial interference noise on the eavesdropper. The relay acts as a traditional relay and retransmits source signal from the source to the intended destination. This dissertation studies the security problem in wireless networks, cooperative networks and wireless communication process of cyber-physical systems (CPS).

In the approaches to friendly jammers, most of references give priority to the demand of source in the approaches of friendly jammers. Friendly jammers are considered as volunteers. However, the jammers play active roles in modern communication but show the selfish character. More precisely, they have no obligations to help the sources’ communication and they are free to choose whether to participate in this system or not. Therefore, friendly jammers will focus the profit of themselves when they participate in the communication system or quit the system when their requirements are not met.

Considering this situation, we study the eavesdropping defense problem in the presence of selfish jammers, who desire to achieve the maximum profit of themselves. In addition, we propose the communication systems including selfish jammers with different prices of per unit power.

However, realizing this process is not trivial because we need to address two challenging research questions: (1) the channel capacity of main channel is poor; and (2) the jammer is not always friendly.

To tackle the two challenges mentioned above, some studies have been done as follows.

For the first challenge, we propose the friendly intermediate nodes which can act as jammers or relays.

We use relays to amplify source signal and retransmit it to intended destination. In this approach, friendly jammers cooperate with relays to defend the eavesdropping attacks. Relays enhance the transmission quality of main channels and friendly jammers disturb malicious eavesdroppers. In this situation, we degrade the wiretap channel and enhance the main channel. To achieve the maximum secrecy capacity by the selected nodes, we introduce a power allocation approach based on price competition.

For the second challenge, we consider the situation that jammers are malicious. There are joint jamming attacks and eavesdropping attacks in our system, but the objectives of these two attackers are different. Specifically, the eavesdropping attackers intend to wiretap the transmission information, but the jammers intend to disturb the transmission in the main channel. Due to the coexistence of jamming attacks and eavesdropping attacks, we may leverage the malicious jammers to disturb the malicious

(12)

- 10 - eavesdroppers.

To tackle the two challenges mentioned above, we employ the game theory to provide a flexible method that studies how the jammers, relays and source interact and cooperate with each other for resource allocation and profit acquisition.

Summary of the Dissertation Review Result

This research investigates the optimal power and price strategy solutions to defend the eavesdropping attacks. This research has three main contributions. The first contribution is a friendly but selfish jammer solution to disturb the eavesdropper. Selfishness is a character, and the jammers will make their profit a top priority. Selfish jammers in this communication system have different prices of per unit power. Bertrand game is used to maximizing the profits of jammers with selfish character. The second contribution studies the anti-eavesdropping issue in wireless cooperative network in the presence of jammers and relays, which selected from these intermediate nodes to achieve the maximum source's utility. Friendly jammers cooperate with relays to defend the eavesdropping attacks. Meanwhile, because of the selfishness of these nodes, the selected nodes have desire to pursue the maximum profit of themselves. Bertrand Game based on price competition is applied to model the relationship between jammers and relays. The third contribution considers the situation that jammers are malicious. There are joint jamming attacks and eavesdropping attacks in this system, but the objectives of these two attackers are different. Hence, the malicious jammers can be utilized to disturb the malicious eavesdroppers. Stackelberg game is used to maximize the profits of source with malicious jammer. These jammers have different prices of per unit power.

The candidate has an excellent scholastic aptitude which is reflected in his strong record of publications and achievements. The candidate has excellent English command. Both review sessions went smoothly. Having carefully evaluated the submitted dissertation by the applicant, the committee unanimously agrees that the contribution of the dissertation is significant to the field of communication networking. In overall, the candidate is fully qualified for the conferment of PhD degree considering the contribution of the dissertation, his publication record and his scholastic ability.

(13)

- 11 -

博士学位論文 Doctoral Dissertation

内容の要旨及び審査結果の要旨 Dissertation Abstract

and

Summary of the Dissertation Review Result

第31号 The Thirty-First Issue

平成30年3月 March, 2018

発行会津大学

〒965-8580 福島県会津若松市一箕町鶴賀 TEL: 0242-37-2600

FAX: 0242-37-2526 THE UNIVERSITY OF AIZU Tsuruga, Ikki-machi Aizu-Wakamatsu City

Fukushima, 965-8580 Japan

博 士 学 位 論 文

Doctoral Dissertation

Dissertation Abstract and

Summary of the Dissertation Review Result

31

The Thirty-First Issue

30

3

March, 2018

The University of Aizu

Preface

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase” or the Doctoral

Degree granted by the University at which the grantee was enrolled.). The Chinese

character,

been granted the degree in accordance with the provisions provided for in Paragraph 4 -2 of

the Ruling Of Degrees (what is called

Contents

Associate

Abstract

Summary of the Dissertation Review Result

Abstract

Summary of the Dissertation Review Result

Abstract

Summary of the Dissertation Review Result

博士学位論文