The Ninth IEEE International Conference on

Awareness Science and Technology (iCAST 2018) September 19 - 21, 2018

&

The Third International Five-Sense Symposium (5-SENSE 2018) September 19 - 20, 2018

Kyushu University, Fukuoka, Japan

Table of Contents

Welcome Message from iCAST 2018 & 5-SENSE 2018 General Chairs …………..….. 1

Welcome Message from iCAST 2018 Program Chairs ……………………………………. 2

Welcome Message from 5-SENSE 2018 Program Chairs ………………………………… 3

iCAST 2018 Organizing Committee …………………………………………………………. 4

iCAST Steering Committee …………………………………………………………………… 4

iCAST 2018 Program Committee …………………………………………………………….. 5

5-SENSE 2018 Organizing Committee …………………………………………………..….. 7

Sponsors ……………………..………………………………………………………..………….. 8

Venue …………………………..……………………………………………….……………..….. 9

Lunch, Reception, and Banquet ……………………………………..…………………………. 11

Overview of Program …………………………………………………………………………….. 12

Plenary Talks ……………………………………………………………………………………… 16

iCAST 2018 Technical Program ………………………………………………………………... 18

iCAST 2018 Paper Abstracts ………………………………………………………………..….. 22

5-SENSE 2018 Technical Program …………………………………………………………….. 42

5-SENSE 2018 Poster Abstracts …………………………………………………………….…. 43

Welcome Message from

iCAST2018 & 5-SENSE 2018 General Chair On behalf of the iCAST 2018 Organization Committee, it is my pleasure to welcome you to the Ninth IEEE International Conference on Awareness Science and Technology (iCAST 2018) held at Kyushu University, Fukuoka, Japan in September 19 - 21, 2018.

Following past eight iCAST conference series, iCAST 2018 was organized to provide a forum for researchers, engineers, and students from all over the world, to discuss the state-of-the-art in awareness science and technology and their applications in wide areas.

The biggest feature of this conference is shown in its conference theme, Awareness Computing Meets Perceptual Psychology. We realized that human science research is necessary for iCAST series to jump up from awareness engineering applications to real awareness science and technology through organizing past iCAST series. For this purpose, we organized the 3rd International Five-Sense Symposium (5-SENSE 2018) at the same venue on the same days, plan joint events, and allow all registrants to attend both conferences freely. We expect that these events provide an opportunity for the researchers to meet and discuss the latest solutions, scientific results and methods in awareness computing and an opportunity to learn something from psychologists.

We would like to express our appreciations to authors who submitted their papers from 19 countries. This wide spread of authors in the world is worthy to the international conference. Thanks to voluntary efforts of referees and Organizing Committee members. We applied a similarity check tool, iThenticate / CrossCheck to all submitted papers and accepted qualified papers with 62% acceptance rate to keep the quality of iCAST 2018 high.

iCAST 2018 and 5-SENSE 2018 were co-sponsored by Japan Chapter of IEEE Systems, Man and Cybernetics Society (SMCS) and Research Center for Applied Perceptual Science (ReCAPS) of Kyushu University, and iCAST2018 is technically supported by SMCS Technical Committee on Awareness Computing. The success of iCAST 2018 was brought by contributions of many volunteers and these organizations. We would like to thank all the authors for submitting their papers, conference attendees for their presentations and discussions, and two plenary speakers, Prof. Rie Matsunaga and Dr. Peter Lewis. We also would like to thank members of our Organization Committee, Program Committee, Publicity Committee, Steering Committee, and staff students for their hard works.

Hideyuki Takagi

General Chair Kyushu University, Japan

1

Welcome Message from iCAST 2018 Program Chairs On behalf of the Program Committee of 9th IEEE International Conference on Awareness Science and Technology (IEEE iCAST 2018), we welcome all participants to joint IEEE iCAST 2018 held at Kyushu University, Ohashi Campus, Fukuoka, Japan in September 19 - 21, 2018. IEEE iCAST 2018 totally received 99 valid submissions from 263 author in 19 countries. Thanks to the strong efforts of 78 International Program Committee members, 61 papers have been finally selected for presentation at the conference. The acceptance rate is 61.6%. Almost all papers have been reviewed by at least three PC members except a few papers with two reviews. The submissions per region were Japan the largest share (36%), Egypt (44%), Taiwan (20%), Indonesia (7%), China (6%), India (5%), and so on. IEEE iCAST 2018 attracted various original works on awareness technologies, and the application supported by awareness technology, such as energy-awareness, robotics, and Bio-related applications. The submission topics covered Sensor (IoT), Big Data, Security and Deep-learning, which are key technologies to support awareness. Meanwhile, we have four special sessions in IEEE iCAST 2018, discussing the promising techniques in awareness, including Biomedical Engineering and Healthcare for Awareness Science and Engineering, Effective Pattern classification from Multimedia Time Series Data for Assessment of Awareness, Intuitive Human - System Interaction, and Intelligent Application of Data Analysis for Content Awareness. We would like to deliver our appreciation to the special session chairs for their hard work. Finally, we would like to thanks Prof. Rie Matsunaga and Dr. Peter Lewis to deliver their Plenary Talks in IEEE iCAST 2018. We hope that we could compile an interesting program for these three days that will appeal all attendees of the conference, and give sufficient opportunities to present newest research, unrevealing new research directions and many plans for future collaborations. Looking forward to meet you all in Fukuoka this autumn.

Junbo Wang Program Chair University of Aizu, Japan

Xin Zhu Program Chair University of Aizu, Japan

2

Welcome Message from 5-SENSE 2018 Program Chairs Welcome to the 3rd International Five-Sense Symposium. We are very happy to hold this

symposium and welcome participants from both inside and outside of Japan.

Since the first symposium in Hita, Japan, the size of this symposium series has

remained small but specialized. Researchers in each of the five senses and with different

backgrounds gather in common sessions and intensively discuss their work and the work

of others from various viewpoints. In this third symposium, we have invited two plenary

speakers for a joint plenary session with the ninth IEEE International Conference on

Awareness Science and Technology (iCAST 2018), and six keynote speakers for the first

and second sessions. Their topics vary greatly, from basic psychophysics to applications in

device engineering. Additionally, seventeen researchers, most of whom are talented

graduate students, will participate in the poster session. We very much hope that

participants of the 5-SENSE 2018 symposium and iCAST 2018 will actively discuss

research concerning the five senses, and that these discussions will stimulate young

researchers and encourage future developments in five-senses research. We would like to

thank the IEEE SMC Japan Chapter for the opportunity to formulate this joint program,

the Research and Development Center for Taste and Odor sensing, Kyushu University,

for organizing the first session, and finally, all the presenters and participants of the

symposium for their fruitful discussions.

Hiroyuki Ito

Program Chair

Center for Applied Perceptual Science

Kyushu University, Japan

3

iCAST 2018 Organizing Committee General Chairs Hideyuki Takagi, Kyushu University, Japan PC Chairs Junbo Wang, University of Aizu, Japan Xin Zhu, University of Aizu, Japan Special Sessions Chairs Hideyuki Takagi, Kyushu University, Japan Publicity Chair Yu-Huei Cheng, Chaoyang University of Technology, Taiwan Publication Chair Kei Ohnishi, Kyushu Institute of Technology, Japan Finance Chair Takashi Yoshida, NEC Corporation, Japan Local Arrangement Chair Kazuo Ueda, Kyushu University, Japan Registration Chair Osamu Maruyama, Kyushu University, Japan Webmaster Jun Yu, Kyushu University, Japan

iCAST Steering Committee Goutam Chakraborty, Iwate Prefecture University, Japan Qiangfu Zhao, University of Aizu, Japan Rung-Ching Chen, Chaoyang University of Technology, Taiwan Robert Kozma, University of Memphis, USA Tadahiko Murata, Kansai University, Japan Kurosh Madani, University of Paris-EST, France

4

iCAST 2018 Program Committee Doi Akio (Iwate Prefectural University, Japan)

Kenji Araki (Hokkaido University, Japan)

Cédric Bornand (HEIG-VD / HESSO, Switzerland)

Oscar Castillo (Tijuana Institute of Technology, Mexico)

Basabi Chakraborty (Iwate Prefectural University, Japan)

Goutam Chakraborty (Iwate Prefectural University, Japan)

Bhabotosh Chanda (Indian Statistical Institute, India)

Bao Rong Chang (National University of Kaohsiung, Taiwan)

Siguang Chen (Nanjing University of Posts and Telecommunications)

Wuhui Chen (Sun Yat-Sen University, China)

Zhe Chen (Dalian University of Technology, China)

Mehdi Darbandi (Iran University of Science and Technology, Iran)

Mingcong Deng (Tokyo University of Agriculture and Technology, Japan)

Harshad Dhotre (University of Bremen, Germany)

Rolf Drechsler (University of Bremen, Germany)

Arnaud Flori (Université Paris-Est Créteil Val de Marne, France)

Asmaa Ghoumari (Université Paris-Est Créteil Val de Marne, France)

Mehran Goli (University of Bremen, Germany)

Masafumi Hagiwara (Keio University, Japan)

Takako Hashimoto (Chiba University of Commerce, Japan) Tomonori Hashiyama (The University of Electro-Communications, Japan)

Fujioka Hiroyuki (Fukuoka Institute of Technology, Japan)

Katsuhiro Honda (Osaka Prefecture University, Japan)

Eenjun Hwang (Korea University, South Korea)

Saori Iwanaga (Japan Coast Guard Academy, Japan)

Christer Johansson (University of Bergen, Norway)

Eisuke Kita (Nagoya University, Japan)

Zbigniew Kokosinski (Cracow University of Technology, Poland)

Setsuya Kurahashi (University of Tsukuba, Japan)

David Lemma (University of Bremen, Germany)

Xiang Li (The University of Aizu, Japan)

Zhi Liu (Shizuoka University, Japan)

Adel M.Alimi (University of Sfax, Tunisia)

Zhenhe Ma (Tianjin University, China)

Takao Maeda (The University of Aizu, Japan)

Yusuke Manabe (Chiba Institute of Technology, Japan)

Konstantin Markov (The University of Aizu, Japan)

Masafumi Matsuhara (Iwate Prefectural University, Japan)

Helmut Mayer (University of Salzburg, Austria)

Subhas Mukhopadhyay (Macquarie University, Australia)

Hien Nguyen (University of Wisconsin Whitewater, USA)

Kazuhiro Ohkura (Hiroshima University, Japan)

5

Kei Ohnishi (Kyushu Institute of Technology, Japan)

Incheon Paik (University of Aizu, Japan)

Jagdish Patra (Swinburne University of Technology, Australia)

Yan Pei (The University of Aizu, Japan)

Anh Pham (The University of Aizu, Japan)

Thuy Pham (Université Paris-Est Créteil Val de Marne, France) Hongzhi Qi (Tianjin University, China)

David Ramamonjisoa (Iwate Prefectural University, Japan)

Keun Ho Ryu (Chungbuk National University, South Korea)

Christophe Sabourin (Université Paris-Est Créteil Val de Marne, France)

Shigeaki Sakurai (Toshiba Corporation, Japan)

Koichi Sato (The University of Aizu, Japan)

Jungpil Shin (The University of Aizu, Japan)

Saeideh Shirinzadeh (University of Bremen, Germany)

Yukari Shirota (Gakushuin University, Japan)

Patrick Siarry (Universit de Paris, France)

Frank Sill (Cyber-Physical Systems of the German Research Center, Germany)

Chunhua Su (The University of Aizu, Japan)

Hideyuki Takahashi (Tohoku University, Japan)

Zunyi Tang (Osaka Electro-Communication University, Japan)

Toshio Tsuji (Hiroshima University, Japan)

Julian Villegas (The University of Aizu, Japan)

Kai Wang (National University of Kaohsiung, Taiwan)

Leon Wang (National University of Kaohsiung, Taiwan)

Yodai Watanabe (The University of Aizu, Japan)

Yutaka Watanobe (University of Aizu, Japan)

Tongquan Wei (East China Normal University, China)

Celimuge Wu (The University of Electro-Communications, Japan)

Yilang Wu (The University of Aizu, Japan)

Minpeng Xu (Tianjin University, China)

Yuichi Yaguchi (University of Aizu, Japan)

Neil Yen (University of Aizu, Japan)

Masao Yokota (Fukuoka Institute of Technology, Japan)

Jun Yu (Kyusyu University, Japan)

Arkady Zgonnikov (The University of Aizu, Japan)

Qiangfu Zhao (The University of Aizu, Japan)

6

5-SENSE 2018 Organizing Committee General Chairs Hideyuki Takagi, Kyushu University, Japan PC Chair Hiroyuki Ito, Kyushu University, Japan Technical Chairs Yoshitaka Nakajima, Kyushu University, Japan Local Arrangement Chair Kazuo Ueda, Kyushu University, Japan Registration Chair Osamu Maruyama, Kyushu University, Japan Webmaster Jun Yu, Kyushu University, Japan

7

Sponsors for iCAST 2018 Co-Sponsors Japan Chapter of IEEE Systems, Man and Cybernetics Society (SMC) Research Center for Applied Perceptual Science, Kyushu University Supporters Technical Committee on Awareness Computing of IEEE SMC Society

Sponsors for 5-SENSE 2018 Co-Sponsors Research Center for Applied Perceptual Science, Kyushu University Japan Chapter of IEEE Systems, Man and Cybernetics Society (SMC)

8

Venue

No.1

No.2

No.

3

No.4

No.5

No.

6

No.7

No.

8

library

admin.

To JR Takeshita Station

To Tenjin

bus

terminal

Technical Sessions: Building #5, Ohashi Campus, Kyushu University 4-9-1, Shiobaru, Minami-Ku, Fukuoka

1 km

9

iCAST 2018 & 5-SENSE 2018 Rooms

524WCWC EV

521

iCAST2018

iCAST2018

531

5-SENSE 2018

poster

WC WC EV

Building #5 3F

Building #5 2F

Building #5 1F

511

WC WC

WC EV

Plenary Talks

10

Lunch, Reception and Banquet Lunches University cafeteria in the building, Design Common, is available for lunch, and there are many restaurants for lunch, convenience stores, bakeries, and supermarkets around the Ohashi Station; see them at http://www.design.kyushu-u.ac.jp/~icast/OhashiLunches.pdf. Reception Reception for all registrants of iCAST2018 and 5-SENSE2018 is held in 18:00 - 20:00 on Sept. 19 (Wed) at the above cafeteria. Your conference name card is an admission ticket. Banquet Banquet is held in 18:30 - 21:00 on Sept. 20 (Thu) at an IZAKAYA-style restaurant bar, Miraizaka (1-8-1, Ohashi, Minami-ku, Fukuoka, Tel: 050-7302-5458). Banquet is not for all registrants of two conferences but only those who ordered a banquet ticket in advance. Banquet includes course dishes and several kinds of alcoholic/non-alcoholic beverages that you may drink as much as you like. Banquet ticket holders are suggested to stay at the Room #511 for a plenary session after the plenary talks end at 18:00. We will guide them to the restaurant at 18:15.

No.1

No.2

No.

3 No.4

No.5

No.

6

No.7

No.

8

library

admin.

Miy

ake

Str

.

bus

terminal

100m

lunch reception

banquet

11

Overview of Program

19 September, Wednesday

Date & Time iCAST2018 5-SENSE

Room 521 Room 524

14:00 - 14:15 Opening

14:20 - 16:00 Awareness

Technology

SS: Biomedical Engineering and Healthcare for

Awareness Science and Engineering

16:00 - 16:20 Break

16:20 - 17:40 Energy-related

Awareness Bio-related Awareness

18:00 - 20:30 iCAST2018 & 5-SENSE2018 Reception

20 September, Thursday

Date & Time iCAST2018 5-SENSE iCAST2018 &

5-SENSE

Room 521 Room 524 Room 531 Room 511

9:20 - 11:00 Security

SS: Effective Pattern classification from

Multimedia Time Series Data for Assessment of

Awareness

Session on Taste and Odor

Sensing

11:00 - 11:15 Break

11:15 - 12:35 SS: Intuitive

Human - System Interaction (1)

Robot-related Applications

Session on Perceptual Psychology

11:35 - 13:50 Lunch

13:50 - 15:30 SS: Intuitive

Human - System Interaction (2)

SS: the Intelligent Application of Data Analysis for Content

Awareness

Poster Session

15:30 - 15:50 Break

15:50- 16:00

Announcement

16:00 - 18:00 Plenary Talks

18:30 - 21:00 iCAST2018 & 5-SENSE2018 Optional Banquet

21 September, Friday

Date & Time iCAST2018

Room 521 Room 524

9:20 - 10:40 Sensor, IoT Web Applications

10:40 - 10:55 Break

10:55 - 12:15 Big Data Deep Learning

12

Room #521 (iCAST2018) Room #524 (iCAST2018)14:00-14:15 iCAST 2018 Opening

14:20 - 16:00 Awareness Technology SS: Biomedical Engineering and Healthcare for Awareness Science and Engineering

14:20-14:40 Pre-accident Situation Analysis Based on Local Motion Menstrual Cycles of Autonomic Functions and Physical Activities14:40-15:00 Towards Systems Awareness A Study on the Development of Portable Wireless Multi-channel

Physiological Signal Measurement System15:00-15:20 Learning Targets for Building Cooperation Awareness in Ensemble

LearningA study on RSVP paradigm based on brain computer interface acrosssubjects

15:20-15:40 Human Awareness Support by Changing Values of Hidden Factors ofInput Stimuli Dynamically

Evaluating the LoRa Wireless Technology for the Application in WaterQuality Monitoring System

15:40-16:00 Evolutionary Problem Solving by People Being Aware of OthersPreferences

Facial skin image classification system using Convolutional NeuralNetworks (CNN) deep learning algorithm

16:00-16:2016:20-17:40 Energy-related Awareness Bio-related Awareness

16:20-16:40 Mapping Technological Trajectories for Energy Storage Devicethrough Patent Citation Network

A Hybrid Brain Computer Interface Driven by Motor Imagery of Right HandVersus Right Forearm

16:40-17:00 Energy Transition, Economic Growth, and CO2 Emission: AnApplication of Energy and Environment in Dynamic Input-OutputModels

Atrial Fibrillation Detection Using Convolutional Neural Network

17:00-17:20 ONU-driven Energy-saving Method in an EPON Implementing Look-ahead MPCP-2

A Facial Pore Aided Detection System using CNN Deep LearningAlgorithm

17:20-17:40 Exploring a Topical Representation of Documents forRecommendation Systems

Segmentation of Lung Nodule in CT Images Based on Mask R-CNN

18:00-20:30

SS: means an organized special session.

(5-SENSE2018)

iCAST2018 & 5-SENSE2018 Program

Break

September 19 (Wed)

iCAST2018 & 5-SENSE2018 Reception (at the Design Common Building on campus)

13

Room #521 (iCAST2018) Room #524 (iCAST2018)09:20-11:00 Security SS: Effective Pattern classification from Multimedia Time

Series Data for Assessment of Awareness09:20-09:40 A New Filter Evaluation Function for Feature Subset Selection with

Evolutionary Computation Marketing Awareness of New Products by Social Network: A Case Study ofHeatTech Products

09:40-10:00 Simple formulation of structural similarity for halftoning and itsapplication to visual secret sharing

Topic-Aware Automatic Snippet Generation for Resolving MultipleMeaning on Web Search Result

10:00-10:20 Activity Strength Recognition Using a Binary Infrared Sensor Array A Single Filter CNN Performance for Basic Shape Classification10:20-10:40 Identifying important factors affecting O2O customers trust from

textual reviewsAspect Aware Optimized Opinion Analysis of Online Product Reviews

10:40-11:00 A Preliminary Experiment on Grid Densities for Visual PasswordFormats

A Proposal for Cost Aware Edge-Detectional Dynamic Time Warping forTime Series Classification

11:00-11:1511:15-12:35 SS: Intuitive Human - System Interaction (1) Robot-related Applications

11:15-11:35 Modeling Non-Compositional Expressions using a Search Engine A human-robot interface to improve facial expression recognition in subjectswith Autism Spectrum Disorder

11:35-11:55 Emoticon-Aware Recurrent Neural Network Model for ChineseSentiment Analysis

Developing a Competency-based System to Enhance KnowledgeManagement Program

11:55-12:15 Estimation of Influence of Each Variable on User’s Evaluation inInteractive Evolutionary Computation

Amoeba Exploration: Coordinated Exploration with Distributed Robots

12:15-12:35 Psychological Effect of Telescope Virtual Screens using VR Headset Design and Implementation of Multiagent-based Evacuation GuidanceSupport System using UAVs

12:35-13:5013:50-15:30 SS: Intuitive Human - System Interaction (2) SS: the Intelligent Application of Data Analysis for

Content Awareness13:50-14:10 Experiment on English-Thai Free Translation via Text Understanding

Based on Mental Image Directed Semantic TheoryUsing Deep Learning to Evaluate the Segmentation of Liver Cell fromBiopsy image

14:10-14:30 Development of Visualization System to Analyze ConversationDocuments in Psychological Counseling

Why Tourists Don’t Visit Again?

14:30-14:50 Single Image Haze Removal Using Weak Dark Channel Prior Some Study of Applying Infra-Red in Agriculture IoT14:50-15:10 Video Summarization: How to Use Deep-Learned Features Without a

Large-Scale DatasetUsing SVM and Random forest for different features selection in predictingbike rental amount

15:10-15:30 Suggestion of the Booting System for Necessary Safety Check byAugmented Reality and Computer Graphics

Leakage-Resilient Certificate-based Encryption Scheme for IoTEnvironments

15:30-15:5015:50-16:0016:00-18:00

16:00-17:0017:00-18:00

18:30-21:00

Plenary Talks (Room #511)

LunchPoster Session

Break

13:50 - 14:30

14:30 - 15:30

flash talks of 19 posters

Poster Presentations

September 20 (Thu)Room #531 (5-SENSE2018)

09:30-10:00

10:00-10:30

10:30-11:00

Taste Sensor and Odor Sensor

Analytical methods for evaluation of food compounds

Collecting of spatial distribution of on-ground odorsources and odor visualization

Peter Lewis, " Self-aware Computing Systems: From Psychology to Engineering "

iCAST2018 & 5-SENSE2018 Optional Banquet (at the Miraizaka Ohashi Restaurant Bar)

Rie Matsunaga, " Cognitive Models of Music Perception "

Announcement (Room #511)

Keynote Session:Taste and Odor Sensing

BreakKeynote Session:

Perceptual Psychology11:15-11:40

11:40-12:05

12:05-12:35

Time-shrinking illusion in the tactile modality:Comparison with auditory and visual modalities

Implementing visual illusions in the real world

On the Origins of the Power Function Exponent

14

Room #521 (iCAST2018) Room #524 (iCAST2018)09:20-10:40 Sensor, IoT Web Applications

09:20-09:40 RFID Impacts on Franchise-Friendly Supply Chain Evaluation of Web Service Recommendation Performance via SparsityAlleviating by Specificity-Aware Ontology-Based Clustering

09:40-10:00 The Assistance for Drug Dispensing Using LED Notification and IRSensor-based Monitoring Methods

A synchronization feedback system to improve interaction correlation insubjects with Autism Spectrum Disorder

10:00-10:20 Design and Implementation of A Caterpillar Robot Which Source Code Plagiarism Detection Approach is More Humane?10:20-10:40 Implementation and Evaluation of VLC-Based Indoor Positioning

Systems for Smart SupermarketsLogic Error Detection Algorithm for Novice Programmers based onStructure Pattern and Error Degree

10:40-10:5510:55-12:15 Big Data Deep Learning

10:55-11:15 Twitter and Online News analytics for Enhancing Post-NaturalDisaster Management Activities

Learning Path Recommender System based on Recurrent Neural Network

11:15-11:35 An Enhanced Hybrid MobileNet A Study on Feature Extraction of Handwriting Data Using Kernel Method-Based Autoencoder

11:35-11:55 Classification of Online Judge Programmers based on Rule Extractionfrom Self Organizing Feature Map

Analysis of Web Service Using Word Embedding by Deep Learning

11:55-12:15 A Data Reconstruction Method for The Big-data Analysis

September 21 (Fri)

Break

15

Plenary Talk 1

Cognitive Models of Music Perception Rie Matsunaga Department of Human Sciences, Kanagawa University, Japan

Abstract Most people associate music with emotional and aesthetic awareness. The emotional awareness of music is not independent from understanding of musical structure. In other words, music perception, at least in part, underlies the emotional awareness of music. In this talk, I will discuss cognitive models of music perception.

As with language, infants are born culture-free with respect to music perception although they may have some innate constraints. As the growing-up process proceeds, listeners’ music perception is largely governed by their musical experiences and learning, or in psychological terms the ‘musical schema’ that reflects implicit knowledge acquired through long-term and everyday exposure to music of their own culture. Thus, when proposing a cognitive model, it should reflect culture-general properties only at a structure level while allowing for flexibility in acquiring and changing internal representations (which correspond to a concept of musical schemas) as a function of music exposure. One important issue is how to represent individual differences of musical schemas in a coherent way while keeping the fixed structure model.

In this talk, after overviewing how listeners perceive a sequence of tones as ‘music’ in their mind, I will focus on a perceptual process of pitch structure or ‘tonality’ perception. Subsequently, I will introduce several behavioral and neuroimaging studies that investigated differences and similarities in tonality perception between listeners of different ages and different music cultures who were at various levels of musical training. Based on evidence of these studies, I will describe fundamental characteristics of cognitive models of tonality perception. Biography Rie Matsunaga obtained Ph. D. degree in Cognitive Psychology from Hokkaido University in 2005. Since April 2017, she is currently an Associate Professor of Department Human Sciences, at Kanagawa University. Prof. Matsunaga's research goal is to understand how mechanisms underlying perception of music arise and change from infancy through adulthood as an experience in one's culture and cognitive (or universal) development processes.

16

Plenary Talk 2

Self-aware Computing Systems: From Psychology to Engineering Peter Lewis ALICE, Aston University, UK

Abstract: There are a number of fundamental changes in the way computing systems are being developed, deployed and used. They are becoming increasingly large, heterogeneous, uncertain, dynamic and decentralised. These complexities lead to behaviours during run time that are difficult to understand, predict, and control. One vision for how to rise to this challenge is to endow computing systems with increased self-awareness, in order to enable more advanced autonomous adaptive behaviour. A desire for self-awareness has arisen in a variety of areas of computer science and engineering over the last two decades.

In this context, we have developed a conceptual framework that provides researchers with a common language with which to describe the self-awareness capabilities of their systems. Our framework is based on a developing fundamental understanding of what self-awareness concepts might mean for the design and operation of computing systems, drawing on self-awareness theories from psychology and other related fields.

I will show how explicit consideration of these concepts may be beneficial in the engineering of adaptive computing systems, that are better able to manage trade-offs between conflicting goals in a complex environment at run time, while reducing the need for a priori domain modelling at design or deployment time.

I will discuss how computational self-awareness may include knowledge of internal state, history, social or physical environment, goals, and further, even a system's own way of representing and reasoning about these things.

Finally, I will describe some of our work in some example application domains: distributed smart-camera networks, volunteer cloud computing, and mobile robotics, where self-awareness can increase runtime adaptivity and robustness, and avoid the need for a priori information at design-time. Biography: Dr. Peter Lewis is a Senior Lecturer in Computer Science at Aston University, in Birmingham. With a background in computational intelligence, Peter has made significant contribution to the field of self-aware computing, including the foundational book Self-aware Computing Systems: An Engineering Approach, in 2016. More broadly, his research is often inspired by biological, social and psychological processes, and advances our understanding about how to build complex computing systems that learn and adapt on an ongoing basis. Through ongoing industrial research collaborations, his work has been applied in areas such as smart camera networks, interactive music devices, avionics, manufacturing, and cloud computing. He obtained his PhD at the University of Birmingham, and is a member of the Aston Lab for Intelligent Collectives Engineering (ALICE).

17

Chair: Paola Di Maio (Center for Technology Ethics, UK)14:20-14:40 Pre-accident Situation Analysis Based on Local Motion 1

Shigeki Kobayashi, Yuichi Yaguchi, Keitaro Naruse, Keita Nakamura and Shuichiro Maekawa14:40-15:00 Towards Systems Awareness 7

Paola Di Maio15:00-15:20 Learning Targets for Building Cooperation Awareness in Ensemble Learning 16

Y. Liu15:20-15:40 Human Awareness Support by Changing Values of Hidden Factors of Input Stimuli Dynamically 20

Hideyuki Takagi, Keisuke Ikeda and Weiqiang Lai15:40-16:00 Evolutionary Problem Solving by People Being Aware of Others' Preferences 26

Ryohei Matsumoto, Momoko Kanmura, Kei Ohnishi and Shinya Watanabe

Session Organizers: Shing-Hong Liu and Prof. Chiun-Li ChinChair: Shing-Hong Liu (Chaoyang University of Technology, Taiwan)

14:20-14:40 Menstrual Cycles of Autonomic Functions and Physical Activities 32Emi Yuda and Junichiro Hayano

14:40-15:00 A Study on the Development of Portable Wireless Multi-channel Physiological Signal MeasurementSystem

36

Shing-Hong Liu, Shao-Heng Lai and Tai-Shen Huang15:00-15:20 A study on RSVP paradigm based on brain computer interface across subjects 42

Yue Sheng, Shuang Liu, Wei Wang, Yuchen He, Xiaoya Liu, Yufeng Ke, Xingwei An and Dong Ming15:20-15:40 Evaluating the LoRa Wireless Technology for the Application in Water Quality Monitoring System 47

Ching-Chuan Wei, Yi-Siang Ciou and Yung-Fa Huang15:40-16:00 Facial skin image classification system using Convolutional Neural Networks (CNN) deep learning

algorithm51

Chiun-Li Chin, Wei-En Chen, Ming-Chieh Chin and Ting-Yu Tsai

Chair: Vimal Kumar (Chaoyang University of Technology, Taiwan)16:20-16:40 Mapping Technological Trajectories for Energy Storage Device through Patent Citation Network 56

Vimal Kumar, Kuei-Kuei Lai, Yu-Hsin Chang and Chien-Yu Lin16:40-17:00 Energy Transition, Economic Growth, and CO2 Emission: An Application of Energy and Environment in

Dynamic Input-Output Models62

Cheng Yih Hong, Yi Chi Tsai and Tsai Rong Lee17:00-17:20 ONU-driven Energy-saving Method in an EPON Implementing Look-ahead MPCP-2 67

Ganbold Shagdar and Buyankhishig Zundui17:20-17:40 Exploring a Topical Representation of Documents for Recommendation Systems 73

Israel Mendonca Dos Santos, Antoine Trouve, Akira Fukuda and Kazuaki Murakami

Chair: Osamu Maruyama (Kyushu University, Japan)16:20-16:40 A Hybrid Brain Computer Interface Driven by Motor Imagery of Right Hand Versus Right Forearm 79

Zhitang Chen, Xin Zhao, Zhongpeng Wang, Kun Wang, Weibo Yi, Feng He and Hongzhi Qi16:40-17:00 Atrial Fibrillation Detection Using Convolutional Neural Networks 84

Xue Zhou, Xin Zhu, Keijiro Nakamura and Mahito Noro17:00-17:20 A Facial Pore Aided Detection System using CNN Deep Learning Algorithm 90

Chiun-Li Chin, Zih-Yi Yang, Rui-Cih Su and Cheng-Shiun Yang17:20-17:40 Segmentation of Lung Nodule in CT Images Based on Mask R-CNN 95

Menglu Liu, Junyu Dong, Xinghui Dong, Hui Yu and Lin Qi

iCAST 2018 Technical Program

Session on Awareness Technology (Room #521)

Special Session on Biomedical Engineering and Healthcare for Awareness Science and Engineering

Session on Energy-related Awareness (Room #521)

proceedingspages

19 September (Wed)

Session on Bio-related Awareness (Room #524)

18

Chair: Qiangfu Zhao (the University of Aizu, Japan)09:20-09:40 A New Filter Evaluation Function for Feature Subset Selection with Evolutionary Computation 101

Atsushi Kawamura and Basabi Chakraborty09:40-10:00 Simple formulation of structural similarity for halftoning and its application to visual secret sharing 106

Ryosuke Kawakubo and Yodai Watanabe10:00-10:20 Activity Strength Recognition Using a Binary Infrared Sensor Array 111

Shoichi Ichimura, Ryo Ota and Qiangfu Zhao10:20-10:40 Identifying important factors affecting O2O customers trust from textual reviews 117

Wan-Ting Chien, Goutam Chakraborty and Long-Sheng Chen10:40-11:00 A Preliminary Experiment on Grid Densities for Visual Password Formats 122

Yesaya Tommy Paulus, Herlina Herlina, Khairu Zeta Leni, Chihiro Hiramatsu and Gerard B. Remijn

Session Organizers: Basabi Chakraborty, Takako Hashimoto, and Goutam ChakrabortyChair: Basabi Chakraborty (Iwate Prefectural University, Japan)

09:20-09:40 Marketing Awareness of New Products by Social Network: A Case Study of HeatTech Products 128Takumi Yoshino, Arisa Takura and Yukari Shirota

09:40-10:00 Topic-Aware Automatic Snippet Generation for Resolving Multiple Meaning on Web Search Result 133Hiroyuki Abe, Masafumi Matsuhara, Goutam Chakraborty and Hiroshi Mabuchi

10:00-10:20 A Single Filter CNN Performance for Basic Shape Classification 139Kenya Murata, Masataka Mito, Daisuke Eguchi, Yuichiro Mori and Masahiko Toyonaga

10:20-10:40 Aspect Aware Optimized Opinion Analysis of Online Product Reviews 144Subha Jyoti Das and Basabi Chakraborty

10:40-11:00 A Proposal for Cost Aware Edge-Detectional Dynamic Time Warping for Time Series Classification 150Hidetoshi Ito and Basabi Chakraborty

Session Organizers: Masao Yokota and Kenji ArakiChair: Masao Yokota (Fukuoka Institute of Technology, Japan)

11:15-11:35 Modeling Non-Compositional Expressions using a Search Engine 155Cheikh M. Bamba Dione and Christer Johansson

11:35-11:55 Emoticon-Aware Recurrent Neural Network Model for Chinese Sentiment Analysis 161Da Li, Rafal Rzepka, Michal Ptaszynski and Kenji Araki

11:55-12:15 Estimation of Influence of Each Variable on User’s Evaluation in Interactive Evolutionary Computation 167Ryohei Funaki, Kentaro Sugimoto and Junichi Murata

12:15-12:35 Psychological Effect of Telescope Virtual Screens using VR Headset 175Masahiro Yamaguchi, Masayo Matsumura, Hikari Shimada and Kenji Araki

Special Session on Effective Pattern classification from Multimedia Time Series Data for Assessmentof Awareness (Room #524)

Special Session on Intuitive Human - System Interaction (1) (Room #521)

20 September (Thu)Session on Security (Room #521)

19

Chair: Kei Ohnishi (Kyushu Institute of Technology, Japan)11:15-11:35 A human-robot interface to improve facial expression recognition in subjects with Autism Spectrum

Disorder179

Yean Han, Yi-Hsiang Ma, Jia-Yeu Lin, Yuya Nishio, Sarah Cosentino, Chiaki Oshiyama and AtsuoTakanishi

11:35-11:55 Developing a Competency-based System to Enhance Knowledge Management Program 185Yujia Wang, Kongjit Chalermpon, Takahiro Hara and Daichi Amagata

11:55-12:15 Amoeba Exploration: Coordinated Exploration with Distributed Robots 191Keisuke Okumura, Yasumasa Tamura and Xavier Defago

12:15-12:35 Design and Implementation of Multiagent-based Evacuation Guidance Support System using UAVs 196Kenta Katayama, Hideyuki Takahashi, Nobuhide Yokota, Kazuya Sugiyasu and Tetsuo Kinoshita

Session Organizers: Masao Yokota and Kenji ArakiChair: Masao Yokota (Fukuoka Institute of Technology, Japan)

13:50-14:10 Experiment on English-Thai Machine Translation via Text Understanding Based on Mental ImageDirected Semantic Theory

202

Rojanee Khummongkol and Masao Yokota14:10-14:30 Development of Visualization System to Analyze Conversation Documents in Psychological Counseling 208

Yasuo Ebara, Tomoya Uetsuji, Minoru Kamata and Koji Koyamada14:30-14:50 Single Image Haze Removal Using Weak Dark Channel Prior 214

Cheng-Hsiung Hsieh, Qiangfu Zhao and Wen-Chang Cheng14:50-15:10 Video Summarization: How to Use Deep-Learned Features Without a Large-Scale Dataset 220

Didik Purwanto, Yie-Tarng Chen, Wen-Hsien Fang and Wen-Chi Wu15:10-15:30 Suggestion of the Booting System for Necessary Safety Check by Augmented Reality and Computer

Graphics226

Kaoru Mitsuhashi

Session Organizers: Rung-Ching Chen and Ruo-Wei HungChair: Rung-Ching Chen (Chaoyang University of Technology, Taiwan)

13:50-14:10 Using Deep Learning to Evaluate the Segmentation of Liver Cell from Biopsy Image 232Shao Kuo Tai and Yi-Shun Lo

14:10-14:30 Why Tourists Don’t Visit Again? 236Long-Sheng Chen, Shu-Cih Tseng and Goutam Chakraborty

14:30-14:50 Some Study of Applying Infra-Red in Agriculture IoT 241Hung-Yu Chien, Yuh-Min Tseng and Ruo-Wei Hung

14:50-15:10 Using SVM and Random forest for different features selection in predicting bike rental amount 246Yi Chen Shiao, Rung-Ching Chen and Wei Hsiang Chung

15:10-15:30 Leakage-Resilient Certificate-based Encryption Scheme for IoT Environments 251Yuh-Min Tseng, Jui-Di Wu, Ruo-Wei Hung and Hung-Yu Chien

Session on Robot-related Applications (Room #524)

Special Session on Intuitive Human - System Interaction (2) (Room #521)

Special Session on the Intelligent Application of Data Analysis for Content Awareness (Room #524)

20

Chair: Gwo-Liang Liao (National Taitung University, Taiwan)09:20-09:40 RFID Impacts on Franchise-Friendly Supply Chain 257

Yung-Fu Huang, Ming-Wei Weng, Rung-Hung Su, Kuei-Kuei Lai and Manh-Hoang Do09:40-10:00 The Assistance for Drug Dispensing Using LED Notification and IR Sensor-based Monitoring Methods 264

Chin-Chuan Han, Hao-Pu Lin, Chao-Hsu Chang, Chang-Hsing Lee, Jau-Ling Shih, Chun-Sheng Hsuand Jen-Chih Chang

10:00-10:20 Design and Implementation of a Caterpillar Robot 268Li-Chun Liao, Yen-Yu Lin, Chen-Yu Huang, Yun-Chen Tsai and Gwo-Liang Liao

10:20-10:40 Implementation and Evaluation of VLC-Based Indoor Positioning Systems for Smart Supermarkets 273Duc Mai and Anh Pham

Chair: Incheon Paik (the University of Aizu, Japan)09:20-09:40 Evaluation of Web Service Recommendation Performance via Sparsity Alleviating by Specificity-Aware

Ontology-Based Clustering279

Rupasingha Arachchilage Hiruni Madhusha Rupasingha and Incheon Paik09:40-10:00 A synchronization feedback system to improve interaction correlation in subjects with Autism Spectrum

Disorder285

Yi-Hsiang Ma, Yean Han, Jia-Yeu Lin, Yuya Nishio, Sarah Cosentino, Chiaki Oshiyama and AtsuoTakanishi

10:00-10:20 Which Source Code Plagiarism Detection Approach is More Humane? 291Oscar Karnalim and Lisan Sulistiani

10:20-10:40 Logic Error Detection Algorithm for Novice Programmers based on Structure Pattern and Error Degree 297Yuto Yoshizawa and Yutaka Watanobe

Chair: Chung-Yen Su (National Taiwan Normal University, Taiwan)10:55-11:15 Twitter and Online News analytics for Enhancing Post-Natural Disaster Management Activities 302

Kuhaneswaran Banujan, Banage T. G. S. Kumara and Incheon Paik11:15-11:35 An Enhanced Hybrid MobileNet 308

Hong-Yen Chen and Chung-Yen Su11:35-11:55 Classification of Online Judge Programmers based on Rule Extraction from Self Organizing Feature Map 313

Intisar Chowdhury and Yutaka Watanobe11:55-12:15 A Data Reconstruction Method for The Big-data Analysis 319

Masataka Mito, Kenya Murata, Daisuke Eguchi, Yuichiro Mori and Masahiko Toyonaga

Chair: Ryohei Funaki (Kyushu University, Japan)10:55-11:15 Learning Path Recommender System based on Recurrent Neural Network 324

Tomohiro Saito and Yutaka Watanobe11:15-11:35 A Study on Feature Extraction of Handwriting Data Using Kernel Method-Based Autoencoder 330

Quan Dang and Yan Pei11:35-11:55 Analysis of Web Service Using Word Embedding by Deep Learning 336

Takeyuki Miyagi, Incheon Paik and Rupasingha Arachchilage Hiruni Madhusha Rupasingha

Session on Deep Learning (Room #524)

Session on Web Applications (Room #524)

Session on Big Data (Room #521)

Session on Sensor, IoT (Room #521)21 September (Fri)

21

iCAST 2018 Paper Abstracts

Pre-accident Situation Analysis Based on Local Motion Shigeki Kobayashi, Yuichi Yaguchi, Keitaro Naruse, Keita Nakamura and Shuichiro

Maekawa Analyzing the video logs captured by drive recorders is important for identifying the possibility of accidents at work. Pre-accident condition analysis using video logs needs to be automated because manual processing requires large amounts of time and effort. Thus, we propose a system for identifying pre-accident situations in a video log captured by the drive recorder on a forklift. Our system focuses on the local motion in each two frames, which is treated as a histogram of the dense optical flow and divided into a lattice. We then estimate local pixel movements to analyze the differences in motion between safe or dangerous scenes. This feature is employed to recognize pre-accident situations using a neural network. Our experimental results showed that we could identify specific pre-accident situations with an accuracy of around 65%, whereas we could recognize nonspecific pre-accident situations with an accuracy of around 98%. This method is useful for actual applications because it can identify the approximate time that a pre-accident situation occurs in the video recorded by a drive recorder. Towards Systems Awareness

Paola Di Maio AI is becoming increasingly powerful and widespread, providing unprecedented technical capabilities with brain computer interfaces and state-of-the-art mind reading computer prototypes. Humans achieved intelligence and consciousness through evolution, yet within a few generations computer science is reproducing and automating these functions. While AI is useful to enhance human cognition exponentially, if decoupled from awareness and responsibility powerful cognition on the outcomes could be unpredictable, yet we do not yet understand the relationship between cognition, intelligence and awareness. Initiatives to promote Ethical AI are proliferating, yet our current understanding of the notion of ethics is limited. Human limitations including limitations of the understanding of the nature of mind may constrain also the effectiveness of the proposed initiatives to mitigate the risks of unresponsive AI. Wisdom traditions, in particular the teachings of the Sakyamuni Buddha, provide guidelines for ethical behaviour. This paper provides an introduction to some critical issues in relation to intelligent systems design as well as a definition of awareness and discusses the correspondence between the eight fold noble path and good practices in systems engineering. It proposes a 'monarchy' as a model for extended awareness, which spans from internal awareness of the technical system to environmental and universal consciousness as a whole. Learning Targets for Building Cooperation Awareness in Ensemble Learning

Y. Liu Negative correlation learning method is to create different individual learners for building a committee machine. In the original version of negative correlation learning, the learning target on a given data point was set to be the same for all the individual learners in the committee. The same learning target could lead the individual learners to become similar to each other if the learning process would be conducted for long. In order to create more different and cooperative individual learners for a committee machine, different learning targets should be set on each learning data for different individual learners in negative correlation learning. In this paper, negative correlation learning with strong and weak learning targets was implemented. On learning each training data, the individual learners could go to the two different learning directions so that there would be little chance for them to become similar

19 September (Wed) 14:20 - 16:00 (Room 521) Session: Awareness Technology

22

even if the long learning process would be performed. Experimental results would show how the two different learning targets would allow the individual learners to become both weak and different in negative correlation learning. Human Awareness Support by Changing Values of Hidden Factors of Input Stimuli Dynamically

Hideyuki Takagi, Keisuke Ikeda and Weiqiang Lai We propose an awareness support system that helps a user to be aware of the reason of his/her evaluations. Based on our proposed definition of human awareness mechanism, we extract hidden factors of input information using an auto-encoder neural networks and implement its decoder part into an awareness support system. The big feature of this system is to let a user change the values of the extracted hidden factors manually and observe the system outputs that change according to the changes of hidden factors. Experimental results using a task of generating facial emotions with 21 human subjects have shown the effectiveness of this approach. Evolutionary Problem Solving by People Being Aware of Others' Preferences

Ryohei Matsumoto, Momoko Kanmura, Kei Ohnishi and Shinya Watanabe Interactive and human-based evolutionary computation methods both enable people to solve a given problem together, but it is hard for us to analyze the processes of the problem solving because people interact with each other nonlinearly in the methods. Therefore, studies of those evolutionary methods are likely to be practical. To make such evolutionary methods involving people more widely used, they need to be traceable and obtain more trust from users. So, in this study, we develop a new traceable evolutionary method involving people. In the method, two or more people produce and evaluate solutions in turn one by one, while being aware of their preferences each other. In addition, assuming two people solve a problem together, we construct not only an experimental system for the method but also a simulation model of the experimental system. Then, we obtain results of experiments by human subjects and simulations and realize from the results that the simulation results assuming completely rational people are different from the experimental ones, in which cooperation beyond rationality among people can occur.

Menstrual Cycles of Autonomic Functions and Physical Activities

Emi Yuda and Junichiro Hayano Although there are many reports on the fluctuation of autonomic nervous function associated with the menstrual cycle, most of them are the comparison between two points during follicular and luteal phases. Also, autonomic nervous functions under free behavior are influenced by physical activities which could be also affected by the menstrual cycle. This study proposed a method using data of physical activity and heart rate variability measured over the total period of a menstrual cycle and analyzed activity-type-specific autonomic nervous functions. This method revealed the fluctuations of autonomic nervous functions with menstrual cycle that were not reported previously. A Study on the Development of Portable Wireless Multi-channel Physiological Signal Measurement System

Shing-Hong Liu, Shao-Heng Lai and Tai-Shen Huang Multi-channel physiological signal measurement systems that are available at the moment are usually wired ones, such as the BioPack MP150 system. The latter, however, is known for its huge size, required connection to alternated current power, lack of an independent data storage unit, and lack of wireless transmission. This study aims to develop a portable wireless physiological signal measurement system that consists of 8 channels. With TI MSP430 F5438A at its microcontroller unit, it had a compact size, lithium battery to supply the needed power, Bluetooth 3.0 data transmission, and built-in 2G flash memory, and the signals was showed on a tablet, smart phone

19 September (Wed) 14:20 - 16:00 (Room 524) Special Session on Biomedical Engineering and Healthcare for Awareness

Science and Engineering

23

or a notebook computer concurrently. Meanwhile, it also supported the extra power supply, ±3V for the other measurement modules. Researcher could use a self-made sensor circuit, and combined with this system to do the ubiquitous healthcare studies. A study on RSVP paradigm based on brain computer interface across subjects

Yue Sheng, Shuang Liu, Wei Wang, Yuchen He, Xiaoya Liu, Yufeng Ke, Xingwei An and Dong Ming

Most visual brain-computer interface (BCI) speller based on the event related potential (ERP) primarily use matrix layouts, and often need patients to complete spelling with moderate eye movement. The fundamental aim of our study is to enhance the perceptibility of target characters by introducing classical rapid serial visual presentation (RSVP) spellers that do not require any eye movement, thereby applying them to the paralyzed patients suffered from oculomotor nerve dysfunction, such as amyotrophic lateral sclerosis (ALS), spinal cord injury, stroke or muscular dystrophy. To test the feasibility of the proposed RSVP paradigm based-BCI, a series of symbols exploded quickly for 20 participants. The flash stimulus on time was 200 ms, and the off time was 100 ms. The effects of sequential letters on target induction with different colors were studied. The P300 component was locked on the target representation by time. The offline classification showed that the average accuracy of choosing the target symbol among 26 possibilities was as high as 90% and above. When calculating the accuracy across subjects under the condition of a certain sample size, the classification rate was changing, up to 68% with the increase of the number of subjects in the sample. The results showed that RSVP speller based-BCI is a promising new model and can be applied to patients with eye movement disorder. Evaluating the LoRa Wireless Technology for the Application in Water Quality Monitoring System

Ching-Chuan Wei, Yi-Siang Ciou and Yung-Fa Huang In recent years, the rise of the IoT(Internet of Things) has drastically reduced the labor costs of traditional industries, including agriculture, aquaculture, and industrial control. In the western coast of Taiwan, many outdoor aquaculture areas have large and empty breeding areas. Deploying a water quality monitoring system using a wired network is not cost-effective. Considering the cost and transmission distance in a variety of LPWAN(Low Power Wireless Area Network), we evaluated these characteristics and thus choose LoRa(Long Range) as a transmission technology in a variety of technologies. Because of the long-distance, low-cost, and low-power advantages, LoRa is proved to be very suitable for this system. Facial skin image classification system using Convolutional Neural Networks (CNN) deep learning algorithm

Chiun-Li Chin, Wei-En Chen, Ming-Chieh Chin and Ting-Yu Tsai The global consumption trend of facial skin care products market is gradually changing. With the concept of preventing aging from becoming more common, the age level of using facial skin care products is gradually reduced, so that the demand of young consumer groups gradually increases. This paper used a deep learning algorithm based on the combination of a smart phone and facial skin detection to develop a facial skin image classification system using Convolutional Neural Networks (CNN) deep learning algorithm. In this system, it can recognize three classes facial skin problem, good facial skin quality, bad facial skin quality and face makeup, and help people quickly understand their facial skin problem. We proposed two different CNN architectures. One has two convolutional layers, two pooling layers and three fully connected layer. The other has three convolution layers, three pooling layers, and four fully connected layer. Finally, we compare the result of our proposed architecture with LeNet-5. From the experimental result, we understand that the architecture which has three convolution layers, three pooling layers, and four fully connected layer, has the highest recognition rete, and we use it as a baseline to build a framework for detecting facial skin problem.

24

19 September (Wed) 16:20 - 17:40 (Room 521) Session on Energy-related Awareness

Mapping Technological Trajectories for Energy Storage Device through Patent Citation Network

Vimal Kumar, Kuei-Kuei Lai, Yu-Hsin Chang and Chien-Yu Lin Technological innovation is a well-known opportunity and it has given the new concept to the competitive era and makes the world easy. It is applied to technological evolution and scientific principles consideration. The innovations in any field occurred where the innovative paths provided by technological trajectories. Further, these technological trajectories are analyzed by the descriptive analysis whereas, the knowledge flow of a technology from one country, institution or corporation to another has been used as a method of identifying by analyzing the patent citation network. It is argued that the quantitative method is used for mapping the technological trajectories and patent citation network. This paper addresses the patent network analysis and its bibliometric analysis is used to monitor the technological trajectories for energy storage device (ESD). The current research focuses on the enhancement and growth of energy storage device. We have focused energy storage device and formed patent citation network from the rough data using USPTO (United States Patent and Trademark Office) database. Further, the scope of this study focuses on key innovation for the energy storage device and its technological trajectories can be considered as patent citation network. Further, it provides the way to develop the main path technology for the industry. Energy Transition, Economic Growth, and CO2 Emission: An Application of Energy and Environment in Dynamic Input-Output Models

Cheng Yih Hong, Yi Chi Tsai and Tsai Rong Lee The year 2011 earthquake in Northeast Japan and the incident of Fukushima nuclear power plant has inspired energy policy in Taiwan to be reevaluated. The Non-nuclear homeland project announced in 2017 has declared a nuclear free energy policy to adjust energy proportion to replace nuclear power by renewable energy before year 2025. This study mainly focuses on the evaluation of the economic effects generated by solar power investment. This study also compared the CO2 reduction contribution of replacing coal-fire electricity by solar power and wind power. The empirical result of our energy and environment in dynamic input-output models revealed the total economic effect induced by renewable energy investment would reach US $95,989.70 million. The CO2 emission could be reduced by 95-98% when replacing coal-fire electricity by solar power and wind power. Such findings have illustrated the economic benefit and CO2 reduction contribution for renewable energy transition. ONU-driven Energy-saving Method in an EPON Implementing Look-ahead MPCP-2

Ganbold Shagdar and Buyankhishig Zundui We propose a simple and effective optical network unit (ONU)-driven energy-saving method in an Ethernet passive optical network implementing a look-ahead-enhanced multipoint control protocol with a parameter 2 (MPCP-2). In the proposed method, with help of MPCP-2, each ONU has the knowledge of next cycle’s transmission before executing current cycle’s transmission. Therefore, each ONU enters power-saving mode between two consecutive upstream transmissions with a self-defined sleep time immediately after the completion of the current cycle’s transmission. In addition, when the sleep time expires, the ONUs can prolong himself the power-saving mode without waking up in every cycle if specific conditions are satisfied and consequently, no sleep-control message between the ONU and the optical line terminal is required. Thus, energy can be saved at ONUs during the idle time between consecutive cycle transmissions or over several cycles by allowing the ONUs enter power-saving mode under ONU’s self-decision. The simulation results demonstrate that the power consumption of ONUs could be reduced by 52.5–56.0% over the entire traffic load.

25

Exploring a Topical Representation of Documents for Recommendation Systems Israel Mendonca Dos Santos, Antoine Trouve, Akira Fukuda and Kazuaki Murakami

In this paper, we address the performance problems inherited when we use word embedding for recommendation. Free-text documents has no structural constructing rules, and are hard to model. Hence, the problem of having an accurate model,that conveys all the important information is a non- trivial problem. We convert the document to a numeric structure using word-embedding and test two document representations: one based in the center of this numeric representation and the other one based on pre-defined set of topics. We build a free text recommendation system and study how the performance, in terms of precision and recommendation time, is affected by both representations. We then vary the number of topics used to represent documents and verify the tradeoffs inherited from having a compact representation. The more compact the recom-mendation, the shorter the recommendation time, however more information is lost in the compactation process. We empirically test different possibilities for the topics and find an optimal point that is 3 times faster than a baseline and almost as accurate as it.

A Hybrid Brain Computer Interface Driven by Motor Imagery of Right Hand Versus Right Forearm

Zhitang Chen, Xin Zhao, Zhongpeng Wang, Kun Wang, Weibo Yi, Feng He and Hongzhi Qi Motor imagery (MI) based BCI is an important technology for the rehabilitation of motor injured. Although it has been developing for a long time, the recognition of MI location with high spatial resolution still faces great challenges. In this paper, we explored the performance of hybrid paradigm used to recognize MI task of right hand versus right forearm. Seven subjects participated in this study, who were required to imagine clenching hand and lifting forearm under MI and hybrid paradigm respectively. MI paradigm asked subjects to only perform the motor imagery tasks, while in the hybrid paradigm, subjects were given electrical stimulation during imagination. Hybrid paradigm requires subjects perform the same tasks in the same way as MI paradigm and not to pay attention to electrical stimulation deliberately. The time-frequency analysis showed that both the ERD and steady-state somatosensory evoked potential (SSSEP) features could be induced during the hybrid paradigm. Classification results show that the mean classification accuracy of the hybrid paradigm reaches 83%, which is significantly higher than the MI paradigm, with an increase around 14%. This indicates that the hybrid paradigm proposed in this paper can effectively improve the spatial resolution of MI location, which can promote MI-BCI system to complete the reach-and-grasp action naturally. Atrial Fibrillation Detection Using Convolutional Neural Networks

Xue Zhou, Xin Zhu, Keijiro Nakamura and Mahito Noro Atrial fibrillation (AF) is the most common cardiac arrhythmia. AF may lead to stroke, heart failure, sudden death and increase the risk of cardiovascular morbidity. Furthermore, AF draws great attention in clinical practice because of its continuously growing prevalence in aging society. The features for AF diagnosis include absolutely irregular RR intervals, and no discernible and distinct P waves. Paroxysmal AF is usually transient and hard to be found in routine health check. Long-term ECG monitoring may raise the sensitivity of AF’s detection. However, the analysis of huge amount of ECG is time and cost consuming. In this study, we propose a method based on convolutional neural networks for the detection of AF. Through validating with MIT-BIH atrial fibrillation database, we get a sensitivity of 98.9%, a specificity of 99.0%, and an accuracy of 99.0%. A Facial Pore Aided Detection System using CNN Deep Learning Algorithm

Chiun-Li Chin, Zih-Yi Yang, Rui-Cih Su and Cheng-Shiun Yang

19 September (Wed) 16:20 - 17:40 (Room 524) Session on Bio-related Awareness

26

Many people are concerned about their facial skin maintenance. Rough pore is one of the facial skin problems which annoyed many people. The size of facial pore is tiny, and it has various shapes. Therefore, it is difficult to recognize facial pore by using traditional image processing. In this paper, we propose an approach based on convolutional neural networks (CNNs) to develop a facial pore aided detection system. We use the LeNet-5 model as our benchmark architecture, and investigate the performance of different depths network on our facial pore dataset. The facial pore aided detection system will help people understand more about their facial skin problems and properly keep their facial skin well. Segmentation of Lung Nodule in CT Images Based on Mask R-CNN

Menglu Liu, Junyu Dong, Xinghui Dong, Hui Yu and Lin Qi Due to the low-quality of CT images, the lack of annotated data, and the complex shapes and unclear contours of lung nodules, existing methods for lung nodules detection only predict the center of the nodule, whereas the nodule size is a very important diagnostic criteria but is neglected. In this paper, we employed the powerful object detection neural network “Mask R-CNN” for lung nodule segmentation which can provide contour information. Firstly, in the case of imbalance between the positive and negative samples proportions, we train classification networks based on block to find a classification network structure suitable for this problem. Then, the classification network structure is used as the backbone network of the image segmentation network—Mask R-CNN, which performs excellently on natural images. Lastly, Mask R-CNN model trained on the COCO data set is fine-tuned to achieve segmentation of pulmonary nodules on CT images. The resulting model achieved desired accuracy on the LIDC-IDRI dataset.

A New Filter Evaluation Function for Feature Subset Selection with Evolutionary Computation

Atsushi Kawamura and Basabi Chakraborty Feature subset selection is an optimization problem to achieve high classification accuracy with low number of features and low compuational cost in the area of pattern classification or data mining. There are various approaches to obtain this. Basically a search algorithm is used with a fitness function either based on intrinsic characteristics of the data, known as filter type, or based on classification accuracy of the classifier used, known as the wrapper type, to find out the optimum feature subset. Both the approaches have respective merits and demerits. Though lots of algorithms are developed so far, none of them works equally well for all the data sets, specially for very high dimensional data sets. In this work a new feature evaluation measure based on the concept borrowed from topic modelling in text processing, has been developed. The proposed measure is used as a fitness function of evolutionary computational search techniques for designing filter type feature subset selection approach. Simulation experiments with various benchmark data sets have been done for assessing the efficiency of the proposed approach in comparison to the popular conventional filter type feature selection algorithms mRMR and CFS. It is found that the proposed approach is better in terms of selecting lesser number of features with comparable classification accuracy. Specially, the proposed algorithms work better for higher dimensional features and can be proved as an effective solution of feature selection for very high dimensional data. Simple formulation of structural similarity for halftoning and its application to visual secret sharing

Ryosuke Kawakubo and Yodai Watanabe The secret sharing (SS) scheme is a cryptosystem which encrypts a secret into multiple pieces, called shares, so that only qualified combination of shares can be employed to recover the secret. The visual secret sharing (VSS) scheme is an example of SS schemes whose decryption

20 September (Thu) 9:20 - 11:00 (Room 521) Session on Security

27

can be performed by a human without any numerical computations. This paper is an attempt to modify the structural similarity (SSIM) index so that the modified one can measure both the structural and tonal similarities simultaneously and then to employ the modified index to give an optimization-based halftoning suitable for the encryption of VSS schemes. Activity Strength Recognition Using a Binary Infrared Sensor Array

Shoichi Ichimura, Ryo Ota and Qiangfu Zhao Smart environments such as smart homes and smart offices have attracted great attention in recent years. Smart home is one solution for senior care in a super-aging society like Japan. Since the smart home is a private space, devices like a video camera and voice recorder cannot be used. The objective of this research is to investigate technologies for constructing privacy-preserving smart home systems. In this paper, we try to use an array of binary infrared sensors to recognize the activity strengths. By activity strength here we mean the speed of a certain action. Because daily-life activities (DLAs) can be considered time sequences of different activity strengths, results obtained in this paper can provide insights about sensor-based DLA recognition. Experimental results show that an array consisting of 15 sensors can provide information for a machine learner to recognize activity strengths well, and the accuracy does not depend on the location of the subject. Identifying important factors affecting O2O customers trust from textual reviews

Wan-Ting Chien, Goutam Chakraborty and Long-Sheng Chen Recently, O2O (Online to Offline/ Offline to Online) business model is growing rapidly. While selecting a service from various options, Wan pointed out that customers' trust is an important issue. To grow business, it is necessary to enhance the customer’s trust on O2O platforms. In previous studies, questionnaire survey data was used to find crucial factors. But gathering survey from customers need manpower and time. Moreover, customers’ comments are restricted by the set of questions, and can not to describe their thoughts explicitly. In this work. We use textual comments by customer as source data to find important factors. The review data obtained over internet contains more positive review than negative. This imbalance in data, when used for classifier, gives high classification score for positive reviews but fail to identify negative reviews. We found, through experiments, that balancing data before factor selection as well as while training the classifier gives highest score overall classification, namely the geometric mean. We used LASSO for feature selection and under-sampling to create balanced data, for training. Results before and after balancing data are shown. The selected features improves the understanding of service provider, where they should emphasize for increased trust and therefore their business. A Preliminary Experiment on Grid Densities for Visual Password Formats

Yesaya Tommy Paulus, Herlina Herlina, Khairu Zeta Leni, Chihiro Hiramatsu and Gerard B. Remijn

Visual passwords are passwords made by selecting a sequence of objects on a screen, such as symbols, pictures, or patterns, either by manual input or eye-gaze-based input. Visual passwords can be useful alternatives to alphanumeric passwords, particularly for authentication on devices in semi-private or public spaces (e.g., on ATMs, laptops, smartphones, or car dashboards). The grid is an essential factor in the use of a visual password, because it can act as a guide for the position of an object and its identification. In this study, we obtained user judgments of 16 different grid densities for three visual password formats. The grid densities were in between 2×2 to 7×7 cells (rows × columns). The participants were asked to judge how easy to use and how safe they thought the grid densities would be, if they would use it for password authentication with eye tracking in a public setting. The results showed that for each visual password format some grid densities were thought to be relatively difficult to use (e.g., a 7×7 grid) or potentially unsafe (e.g., a 2×2 grid). The range of grid densities in between 3×3 to 6×6 cells, however, is potentially suitable for the three visual password formats studied here, for use with real eye tracking in a public setting.

28

Marketing Awareness of New Products by Social Network: A Case Study of HeatTech Products

Takumi Yoshino, Arisa Takura and Yukari Shirota Marketing awareness has become an important research theme in text mining application fields. Marketing persons are concerned about reputations on SNS concerning their new products. To gauge the reaction of the market, they use sentiment analysis of text mining. However, it is difficult to identify the reaction as positive or negative ones just by using sentiment analysis. In the paper, we shall propose to use stock prices as the measurement index. Because stock prices are a result of many data concerning a product, if a company starts to sell a new main product, the stock prices are likely to reflect the reputation of the newly sales product. If the stock price increases, we can think that the new product are welcome in the market. In the paper, we shall propose a detection model for the clear-cut correlation between a SNS spike and stock price movement. If we find a SNS spike, firstly, topic extraction is conducted on the SNS text data to remove the noise data to extract a purely breaking topic. Then, from the breaking topic distribution and period, we make the differential equation. Finally, we determine whether the solution data matches the actual stock price data. If we find the correlation between the SNS spike and the stock price change, we can predict the future stock price movement. Topic-Aware Automatic Snippet Generation for Resolving Multiple Meaning on Web Search Result

Hiroyuki Abe, Masafumi Matsuhara, Goutam Chakraborty and Hiroshi Mabuchi In recent years, the amount of information on the Web is growing exponentially with the spread of the Internet. We generally use search engines to search for the intended information. However, the search engine in displays the Web pages including the entered search query in list format. It is difficult for the user to find out the intended information if the entered search query is a word whose meaning depends on the situation and location of the user. It needs, the intended information hidden, in the multiple irrelevant topics. In this research, we classify Web search results based on each topic. The topic is defined as the latent meaning, and the contents included in the word. Moreover, our system displays automatically generated snippets for each topic with the Web search results to the user. It is easy to find required information from snippets, even though the intended information is ambiguous. It first classifies the Web search results by Latent Dirichlet Allocation(LDA) which is a major topic model method. It then generate the snippets using Conditional Variational AutoEncoder(Conditional VAE) based on the clustering of We search results. It is expected that using LDA for the clustering will group the Web search result according to the latent meanings of the search query. Also, we expecte that proper snippets will be generated for each topic by Conditional VAE. In this paper, we show that LDA is effective for the clustering of Web search results. Moreover, the snippets generated by Conditional VAE is able to generate sentences considered each topic. A Single Filter CNN Performance for Basic Shape Classification

Kenya Murata, Masataka Mito, Daisuke Eguchi, Yuichiro Mori and Masahiko Toyonaga IoT cameras and sensors collect any images and any sensing data from everywhere in the world to send them via the internet. Those collected images are stacked into the servers, and an image recognition system on the server such as CNN (Convolutional Neural Net) mines valuable knowledge. In the near future, when the enormous number of IoT collects images at various places, these servers would be over-flow to do. So if IoTs would send not only images but also analyzed results to the server, that would reduce the server loads, however, the conventional CNN is too large to implement in it. We propose a single filter CNN model that can be implemented even on a small IoT. Our CNN model is a minimal configuration with an input layer, an affine transformation layer, a convolution layer, a pooling layer and a full

20 September (Thu) 9:20 - 11:00 (Room 524)

Special Session on Effective Pattern classification from Multimedia

Time Series Data for Assessment of Awareness

29

connection layer. We evaluate our proposed CNN model by two experiments. One is to check 1) whether it can learn the eleven basic shapes or not, i.e. a circle, a triangle, a square, and so on. Second is to check 2) whether it can classify the basic shapes against to their shape reduction and their noise mixture. Results for the first experiments, it is found that our system can classify the all basic shapes perfectly. Results for the second experiments shows the accuracy depends on types of filters both for the scaled shape classification and the inverse pixel noise shape classification. Aspect Aware Optimized Opinion Analysis of Online Product Reviews

Subha Jyoti Das and Basabi Chakraborty Now-a-days social media and micro blogging sites are the most popular form of communication. The most useful application on these platforms is Opinion mining or Sentiment classification of the users. Here in this work an automated method has been developed to analyze and summarize opinions on a product in a structured, product aspect based manner. In this way future potential buyers do not need to go through all the reviews manually and can get an idea from easily comprehensible representation of the review data. A Proposal for Cost Aware Edge-Detectional Dynamic Time Warping for Time Series Classification

Hidetoshi Ito and Basabi Chakraborty Dynamic Time Warping (DTW) is a well known algorithm for measuring similarity of two time series and widely used in classification, clustering or regression problems related to time series data. Unlike simple Euclid distance measure DTW can handle time series of unequal lengths and is able to find an optimal alignment between two time sequences. Though very efficient, the computational cost of DTW is very high. There are several suboptimal variants of DTW for lowering computation, none of them is perfect. In this work, an approach to reduce computational burden of DTW has been proposed from the perspective of removing unimportant portion of the time series from computing, selected by a mask generated by edge detection algorithm commonly used in image processing or computer vision. The proposed Edge-Detectional Dynamic Time Warping (EDDTW) has been compared with original DTW by simulation experiments with 43 publicly available benchmark data sets. The simulation results show that EDDTW outperforms DTW regarding classification accuracy in more than half of the data sets, while reducing on the average 60% of the original time series leading to reduction in computational time.

Modeling Non-Compositional Expressions using a Search Engine

Cheikh M. Bamba Dione and Christer Johansson Non-compositional multi-word expressions present great challenges to natural language processing applications. In this paper, we present a method for modeling non-compositional expressions based on the assumption that the meaning of expressions depends on context. Therefore, context words can be used to select documents and separate documents where the expression has different meanings. Deviation from a baseline is measured using serendipity (i.e. the pointwise effect size). We used this statistical measure to mark which patterns are over- and under-represented and to take a decision if the pattern under scrutiny belongs to the meaning selected by the context words or not. We used the Google search engine to find document frequency estimates. When used with Google document frequency estimates, the

20 September (Thu) 11:15 - 12:35 (Room 521)

Special Session on Intuitive Human - System Interaction (1)

30

serendipity measure closely mirrors some human intuitions on the preferred alternative. Emoticon-Aware Recurrent Neural Network Model for Chinese Sentiment Analysis

Da Li, Rafal Rzepka, Michal Ptaszynski and Kenji Araki Pictograms (emoticons/emojis) have been widely used in social media as a mean for graphical expression of emotions. People can express delicate nuances through textual information when supported with emoticons, and the effectiveness of computer-mediated communication (CMC) is also improved. Therefore it is important to fully understand the influence of emoticons on CMC. In this paper, we propose an emoticon polarity-aware recurrent neural network method for sentiment analysis of Weibo, a Chinese social media platform. In the first step, we analyzed the usage of 67 emoticons with facial expression used on Weibo. By performing a polarity annotation with a new "humorous type" added, we have confirmed that 23 emoticons can be considered more as humorous than positive or negative. On this basis, we applied the emoticons polarity in a Long Short-Term Memory recurrent neural network (LSTM) for sentiment analysis of undersized labelled data. Our experimental results show that the proposed method can significantly improve the precision for predicting sentiment polarity on Weibo. Estimation of Influence of Each Variable on User’s Evaluation in Interactive Evolutionary Computation

Ryohei Funaki, Kentaro Sugimoto and Junichi Murata Recently, interactive evolutionary computation (IEC) has been extensively applied in those systems that recommend objects, such as images and sounds, to users based on their preference. If an IEC user's evaluation criteria are clearly known, they can be utilized for acceleration of IEC, merchandise development, and creativity support for designers. It is difficult to collect a large volume of evaluation data for their analysis because an IEC user cannot repeat the evaluation so many times. Therefore, the technique proposed in this study adopts paired comparison-based interactive differential evolution (IDE) to ease the burden of users, and it will extract the user evaluation criteria through less number of evaluation steps. These techniques estimate the user evaluation criteria using the distribution of solutions because IDE does not receive the evaluation values from its user. Techniques are proposed that estimate, through the IEC processes, the degree of influence of each variable on the evaluation by any given user. During the simulations, the proposed methods are evaluated on test problems. Psychological Effect of Telescope Virtual Screens using VR Headset

Masahiro Yamaguchi, Masayo Matsumura, Hikari Shimada and Kenji Araki We carried out a psychological experiment using Virtua Reality Head Mount Display (VR-HMD) with small FOV (Field of view) camera and a screen in virtual environment, telescope virtual screen, to analyze the change of subjects’ pain tolerance. We manipulate FOV of the virtual camera in VR to change the view of contents. Decreasing FOV makes view of angle narrow, therefore the view in VR looks similar to the one using telephoto lens, in which environment the view greatly moves with even small camera movement. With smaller value of FOV for a virtual camera, the view in VR is more sensitive to VR-HMD movement and requires subjects to keep concentrating not to move their heads to watch target properly. Hand immersion into cold water is a standard test for pain tolerance evaluation and is known as the cold pressor test. Cold pressor test with the VR-HMD shows that pain tolerance was increased with telescope virtual screen environment. The result indicates that pain tolerance can be controlled with VR-HMD without meaningful contents. The contents independent pain management technique can potentially be used in various applications especially in medical field.

31

A human-robot interface to improve facial expression recognition in subjects with Autism Spectrum Disorder

Yean Han, Yi-Hsiang Ma, Jia-Yeu Lin, Yuya Nishio, Sarah Cosentino, Chiaki Oshiyama and Atsuo Takanishi

This research integrates WAS-5 the saxophone playing music robot with the music therapy of subjects with Spectrum Disorder (ASD). Robots have been found to cause less anxiety in children of autistic spectrum while music boosts their concentration and learning capability thus music therapy is common for ASD therapy sessions. Researches on the effect of robots on ASD subjects have been lacking on robots which are of human size and plays musical instrument, as most focus on interaction only, thus WAS-5 can be used for such an observation. WAS-5 will be the medium the children interact with, which by using its expressions along with music cues learnt in past music therapy by the subjects, enabling WAS-5 to condition the ASD subjects to differentiate expressions through repetitive training. The goal is to train ASD subjects to learn the social ability to differentiate varying expressions from one another. This paper evaluates the gaze detection system and the training protocol to be used in the therapy. The preliminary experiment of the training protocol has shown positive training effects in normal subjects. Developing a Competency-based System to Enhance Knowledge Management Program

Yujia Wang, Kongjit Chalermpon, Takahiro Hara and Daichi Amagata Properly managing knowledge workers greatly impacts and influences the contemporary globalized world. Although, much research has emphasized what kinds of competency knowledge workers should have, it is difficult to reach a consensus on what constitutes a competent lecturer in the knowledge management (KM) domain. Due to the importance of lecturers’ competencies in the context of the student quality assurance, industry requirements, existing problems in the management process and operational functions in changing business environments, it is necessary to design and validate an effective competency model and implement it into practical systems. In this paper, we develop a competency-based knowledge management system which employs a database and competency model. That is, based on a KM lecturers' competency model, we set up a database so that the internal management process can be easily handled (e.g., administrators can assign the lecture for a particular class, and students can find their advisors). This system demonstrates how a competency model practically supports decision-making and work processes. Amoeba Exploration: Coordinated Exploration with Distributed Robots

Keisuke Okumura, Yasumasa Tamura and Xavier Defago Autonomous exploration with distributed robots, such as search and rescue, planetary exploration and patrolling, is a major challenge for multi-robot coordination. In this paper, we propose a fascinating framework for coordinated exploration, named, amoeba exploration. The aim of amoeba exploration is to take the speed balance between acquiring and sharing information when using robots with limited communication ability. To achieve this purpose, robots have to obey a simple rule; gathering at the dynamic meeting place, called "Anchor", on at regular intervals. Simulation experiments were conducted in several conditions to evaluate our proposal, and amoeba exploration successfully shortened the exploration time. Design and Implementation of Multiagent-based Evacuation Guidance Support System using UAVs

Kenta Katayama, Hideyuki Takahashi, Nobuhide Yokota, Kazuya Sugiyasu and Tetsuo Kinoshita

Huge Disasters such as big earthquakes, tsunamis, typhoons and eruptions occur all over the

20 September (Thu) 11:15 - 12:35 (Room 524)

Session on Session on Robot-related Applications

32

world, so system for disaster prevention is expected. Especially, tsunamis by big earthquake cause serious damage. We propose an evacuation guidance support using multi-UAVs and aim to decrease the damage by tsunamis. Specifically, software agent decide the safest evacuation guidance route for UAVs based on the disaster situations, the elevation of evacuation route and the distance from the route to coastline etc. In addition, the agent changes the route flexibly according to the disaster situation while evacuation guidance. Also, we aim to perform evacuation guidance support effectively by the function of efficient evacuation guidance and helping the UAV that does not work, that is realized by cooperation of multi-UAVs. In this paper, we describe the approach to decide the evacuation guidance route for UAV and change the route flexibly. In addition, we describe the function of cooperation of multi-UAVs. We performed experiments on path re-planning according to disaster situation in simulation and flight of UAV in real world to confirm the effectiveness of our approach.

Experiment on English-Thai Machine Translation via Text Understanding Based on Mental Image Directed Semantic Theory

Rojanee Khummongkol and Masao Yokota This paper describes a methodology for cross-language paraphrase via intermediate semantic expression in a knowledge representation language called Mental-image Description Language (Lmd) and its application to an experimental system for English-Thai free translation. This system interprets English text into Lmd expression to understand it and interprets the understanding result into Thai text without using any syntactic information of the input English text. It works as one kind of inter-lingual machine translation system but actually is a subsystem of our natural language understanding system to paraphrase an input text in another language. Some experimental results were evaluated by several native Thai speakers with good knowledge of English, which gave a good perspective to our future work on this system. Development of Visualization System to Analyze Conversation Documents in Psychological Counseling

Yasuo Ebara, Tomoya Uetsuji, Minoru Kamata and Koji Koyamada In psychological counseling, beginner counselors tend to force the client the interpretation by the counselor himself, the amount of utterance by the beginner conselor become larger than that of the client and use the closed-ended question at high rates to confirm the image of the client created with their own interests. The opportunity of education is referred to as the supervision that expert counselors as the supervisor advice forbeginner counselor is provided. However, the verbatim record of conversation documents in counseling used in the supervision are large-scale and complex, the supervisor is very difficult to extract the characteristics and situation of conversation between the beginner counselor and the client. To improve the problem, we proposed the system for visualizing the conversation documents in the psychological counseling. We evaluated to use this system by some counselors to verify the effectiveness of this proposed system. From this evaluation, the results were obtained from most estimators that these visualization results have a high tendency to be easily understandable. Moreover, we conducted the comparative evaluation by working the same task in the case of reading only the original text and in the case of using this proposed system by using the conversation documents in actual counseling. As the results, we showed that the evaluation results that the task using this proposed system is easier to understand in all of the evaluation items.

20 September (Thu) 13:50 - 15:30 (Room 521)

Special Session on Intuitive Human - System Interaction (2)

33

Single Image Haze Removal Using Weak Dark Channel Prior Cheng-Hsiung Hsieh, Qiangfu Zhao and Wen-Chang Cheng

Haze removal or dehazing has been a challenge in the field of image restoration. Recently, He et al proposed a single image dehazing scheme based on an interesting statistical prior called dark channel prior (DCP). By the DCP, two parameters in the haze image model, the atmospheric light and the transmission map, can be estimated easily. Consequently, the DCP scheme has attracted much attention in this field. Note that the DCP scheme relies on the block-based dark channel which is considered as a strong DCP assumption. In this paper, a pixel-based dark channel is introduced through which the atmospheric light and the transmission map are estimated. The pixel-based dark channel is considered as a weak DCP (WDCP) since its statistical property is not as strong as that in the block-based dark channel. With a similar manner in the DCP scheme, the atmospheric light is estimated through the pixel-based dark channel. To make the pixel-based dark channel feasible in the transmission map estimation, an adaptive scaling factor for the initial transmission map is employed and the pixel-based dark channel is applied as the guide image in the transmission map refinement by the guided image filtering. Furthermore, an objective assessment is used to evaluate the proposed WDCP scheme and the compared dehazing schemes. Simulation results indicate that the proposed WDCP scheme is more efficient, 24.30 times faster than the DCP scheme on average. Moreover, the proposed WDCP scheme is of better subjective visual quality than the DCP scheme and the employed objective assessment agrees with the results in the given examples. Video Summarization: How to Use Deep-Learned Features Without a Large-Scale Dataset

Didik Purwanto, Yie-Tarng Chen, Wen-Hsien Fang and Wen-Chi Wu This paper proposes a framework incorporating deep-learned features with the conventional machine learning models within which the objective function is optimized by using quadratic programming or quasi-Newton methods instead of an end-to-end deep learning approach which uses variants of stochastic gradient descent algorithms. A temporal segmentation algorithm is first scrutinized by using a learning to rank scheme to detect the abrupt changes of frame appearances in a video sequence. Afterward, a peak-searching algorithm, statistics-sensitive non-linear iterative peak-clipping (SNIP), is employed to acquire the local maxima of the filtered video sequence after rank pooling, where each of the local maxima corresponds to a key frame in the video. Simulations show that the new approach outperforms the main state-of-the-art works on four public video datasets. Suggestion of the Booting System for Necessary Safety Check by Augmented Reality and Computer Graphics

Kaoru Mitsuhashi The safety system is recently developing using microcontroller and sensing, such as the safety mode of the numerical control (NC) for machining center, the light curtain for press machine etc. However, the manual control device/tool cannot be developed, because integrating the sensor or the microcontroller is difficult. In this paper, we suggest the booting system for necessary safety check using AR (Augmented Reality) technology. First, the safety check process is constructed as a reference of PDCA using programing. After that, we confirm the validity of the system for safety check using AR technology, the safety check function of the machine tool and switches is investigated by some operators. The machine tool and many switches are equipped with AR markers, and the operators have the web camera on head side. Therefore, the switches are measured correctly, and the operators can check the safety using web camera and AR technology.

34

Using Deep Learning to Evaluate the Segmentation of Liver Cell from Biopsy image

Shao Kuo Tai and Yi-Shun Lo Liver cancer is one of the most critical health problems in the world. The grading diagnosis for liver cancer in biopsy images is essential for the treatment of liver cancer and disease prognosis. A grading system that uses artificial intelligence to provide quantitative and objective results for physicians and pathologists; it will not only save time but also improve the accuracy of diagnosis. In the grading system, the main work is grading with the nucleus segmented from cancer biopsy images. However, improper focus and complex stroma background will affect the performance of segmentation. If we can evaluate segmentation of the nucleus and exclude the failed segmentation from the grading system, it will significantly improve the accuracy of the grading. In this paper, we propose a method with deep learning for evaluating the segmentation of liver nucleus, and the experimental results demonstrate that the performance of our method is 90.5%. Why Tourists Don’t Visit Again?

Long-Sheng Chen, Shu-Cih Tseng and Goutam Chakraborty The population of tourists has grown rapidly with the development of social media. The rise of social media has changed the behaviors that passengers visit sightseeing spots. Online consumer reviews could be considered as the main channel for providing valuable information to consumers. Revisit intention could directly influence the future behavior of customers. It’s also one of the crucial factors that enhance the income growth of tourism. However, relatively few researchers focused on why passengers don’t revisit directly. Therefore, this study will focus on the topic that why passengers didn’t revisit again. We’ll use textual reviews of social media instead of questionnaire survey. And text mining and feature selection (Least Absolute Shrinkage and Selection Operator, LASSO) methods have been employed to identify the factors that affect passenger non-revisit intention. From the results, this study will provide some suggestions for the tourism industry to improve their service quality and increase their revisit intentions. Some Study of Applying Infra-Red in Agriculture IoT

Hung-Yu Chien, Yuh-Min Tseng and Ruo-Wei Hung Applying Internet-of-Things (IoT) technologies in agriculture not only can reduce the man efforts but also improve the productivity and the efficiency. Through the IoT technologies, one can collect various data like luminosity, temperature, humidity, PH value, etc to analyze and control the facilities. In this study, we focus on exploring the application of the infra-red data on agriculture IoT systems, in addition to the design of our green-house IoT system. Through the infra-red data, we can analyze various biological data of the plants and the fruits. Based on several low-cost devices (like Raspberry pi, NoIR camera, thermal camera, various sensors), some open source platforms, and the Splunk software, we design and build an green-house IoT system that not only collects various environment data but also analyze the biological data of plants via the infra-red data. Some preliminary experiments and analysis are given in this paper. The results show that low-cost infra-red devices could have a great potential contribution to improving agriculture practice. Using SVM and Random forest for different features selection in predicting bike rental amount

Yi Chen Shiao, Rung-Ching Chen and Wei Hsiang Chung Nowadays, people rely on bike renting service for transportation in short distance to replace walking. It is more convenient and faster for people to transfer from place to place. Public transportation is very popular for people to go to work or school. However, there might not be

20 September (Thu) 13:50 - 15:30 (Room 524)

Special Session on the Intelligent Application of Data Analysis

for Content Awareness

35

so many stations to let everyone arrive at the place where they want to go. If it takes too much time from stations to destination, it will make people have less willingness in taking public transportations. Bike renting system like U-bike solves this problem. The need for bike renting leads to a question of setting bike rental locations and the number of bikes in each place, by predicting the number of people renting bikes in each position can make it easier for governments to assign bikes to each position. When predicting the bike rent amount, there are lots of features to consider with, like the weather, time, holiday. Using more features doesn't mean to be better, so the selection of the feature is essential. In this paper, we proposed a method which will combine random forest and support vector machine to predict the bike rental amount from the last hour. Experiments results will discuss random forest, super vector machine and the combination of the two methods results. Leakage-Resilient Certificate-based Encryption Scheme for IoT Environments

Yuh-Min Tseng, Jui-Di Wu, Ruo-Wei Hung and Hung-Yu Chien Now, Internet of Things (IoT) brings people innovative experiences and applications through connectivity of numerous computing devices. In these applications, computing devices generate and exchange a large number of critical and sensitive data. Typically, these computing devices are putted on some unprotected environments that make them to be attractive attack targets while easily suffering from a new kind of threat, called “side-channel attacks”. By side-channel attacks, an adversary could obtain partial information of secret values (or internal states) stored in these devices by observing execution timing or energy consumption. However, most adversary models of previous cryptographic schemes/protocols do not concern with such side-channel attacks. Indeed, leakage-resilient cryptography is a flexible solution for resisting to side-channel attacks. So far, little work focuses on the design of leakage-resilient certificate-based encryption (LR-CBE) schemes. In the article, we propose the first LR-CBE scheme resilient to continuous key leakage of user’s private keys, system secret key and random values. In the generic bilinear group model, security analysis is given to show that the proposed LR-CBE scheme is provably secure against chosen cipher-text attacks under the continual leakage model. Performance evaluation is made to demonstrate that our scheme is suitable for embedded devices.

RFID Impacts on Franchise-Friendly Supply Chain

Yung-Fu Huang, Ming-Wei Weng, Rung-Hung Su, Kuei-Kuei Lai and Manh-Hoang Do International franchisers have been generally regarded as market seekers who exploit their firm-specific capabilities overseas. The use of radio frequency identification (RFID) technology for international franchisers plays a key role in improving service reliability and satisfaction. This paper focuses on filling the gap in theoretical and empirical knowledge of inventory management practices of franchise-friendly supply chain in Taiwan. A real case in Taiwan franchise industry is quoted to build a corresponding two warehouse inventory model. The objective is to examines the total cost that can be achieved by one frozen central kitchen (FCK) and multi-branch through RFID technology in franchiser system. The Assistance for Drug Dispensing Using LED Notification and IR Sensor-based Monitoring Methods

Chin-Chuan Han, Hao-Pu Lin, Chao-Hsu Chang, Chang-Hsing Lee, Jau-Ling Shih, Chun-Sheng Hsu and Jen-Chih Chang

In this study, an assistant system is developed for pharmacist to improve the dispensing quality by two functions: notification in time and monitoring in real time. During drug dispensation, the system gets the patient’s prescription issued from doctors, and drives the

21 September (Fri) 9:20 - 10:40 (Room 521)

Session on Sensor, IoT

36

LEDs for notification. Since some drug titles, shapes, colors, or packages are very similar, pharmacists waste lots of time to find the correct drugs. With LED notification, pharmacists pick up the drugs from the correct cabinets, and save the dispensing time. Second, the system monitors pharmacist actions by the IR-sensors. An alarm is given if pharmacists pick up the incorrect drugs or lost the drug items, even the correct LEDs are turn on. In addition, a web-based information system is designed for drug dispensing and inventory management. During the dispensation, patient information and drug data are displayed on the screen for notification. Design and Implementation of a Caterpillar Robot

Li-Chun Liao, Yen-Yu Lin, Chen-Yu Huang, Yun-Chen Tsai and Gwo-Liang Liao This paper proposed a low cost worm-shape robot that can mimic the locomotion of a caterpillar to crawl by arching and stretching its body. Based on the process of implementing the caterpillar robot, the participated students can built up their interdisciplinary skills. Students have to integrate the technologies of micro-controllers, sensors, materials and mechanisms. We used a Motoduino U1 module to drive 4 servo-motors and Bluetooth module to communicate between robot and users. In order to imitate gaits of caterpillars, the body of the robot was made by using discrete PVC rings to flexibly arch. The robot also can be remotely controlled to crawl forward/back or right/left, and change its height of arched back. The weight and total length of the robot were bout 470g and 65cm, respectively. The experiment shows that when the caterpillar robot moves in line, the averaged speed of robot can be 25m/hr. The maximum rotating angle is 30º and the minimum rotating diameter is 120cm. Implementation and Evaluation of VLC-Based Indoor Positioning Systems for Smart Supermarkets

Duc Mai and Anh Pham This paper presents an experimental implementation and performance evaluation of an indoor positioning system based on visible light communications (VLC). The system hardware, including a VLC transmitter and a receiver, is designed taking into account the illumination flicker, brightness, signal synchronization, noise and path loss issues. In addition, we propose a simple protocol design for location identification (ID), including the location ID format, the ID frame structure and its error control. The experiments are performed to optimize the design of the receiver so that the pulse-error rate (PER) is minimized. Furthermore, we also confirm the location ID detectability in case of with and without Golay error correction code. Experimental results confirm that it is possible to achieve the PER bellow 10-3 at 9 m or shorter distance. In addition, the location ID detectability is higher than 95¥% when the receiving time of 0.7 seconds, which is equivalent to the walking speed of 1.4 m/sec, under the LED coverage cell of 2 m.

Evaluation of Web Service Recommendation Performance via Sparsity Alleviating by Specificity-Aware Ontology-Based Clustering

Rupasingha Arachchilage Hiruni Madhusha Rupasingha and Incheon Paik With the development of information technology, considerable web services are published on the Internet rapidly in the last few years. It becomes a challenging task to recommend applicable web services to users and service recommendation becomes an influential approach to guide users to discover suitable services. In this situation, Collaborative Filtering (CF) based on rating is one of the powerful approaches for service recommendation but suffers from the data sparsity and cold-start problems due to the insufficiency of user-service information. In this paper, we present a novel ontology-based clustering approach that based on the terms specificity and similarity to overcome those limitations. We alleviate the sparsity problem

21 September (Fri) 9:20 - 10:40 (Room 524)

Session on Web Applications

37

using this novel clustering approach and then service user similarity is calculated using a Pearson Correlation Coefficient (PCC) measurement. Finally, user rating is predicted based on the alleviated user ratings and PCC values and recommendation is based on these predictions. We evaluated in the several viewpoints based on our previous work and results show that our approach can successfully alleviate the sparsity and cold-start problems and works effectively by lower prediction error compared with existing approaches. A synchronization feedback system to improve interaction correlation in subjects with Autism Spectrum Disorder

Yi-Hsiang Ma, Yean Han, Jia-Yeu Lin, Yuya Nishio, Sarah Cosentino, Chiaki Oshiyama and Atsuo Takanishi

WAseda Saxophonist No.5 (WAS-5) is a saxophone playing robot that will be used in this research to improve the social interaction skills of ASD subjects with other individuals. One interaction skill is to match the tempo of other individuals in talking speed and other physical activities. WAS-5 will be playing music at a certain tempo which the ASD subject are expected to match which WAS-5 will adapt accordingly to the feedback tempo of the movement of the ASD subject through PID control. The movement of the subjects will be captured and processed with a camera. Through the repetition of the experiment, the subjects are expected to be able to grasp the tempo WAS-5 and eventually match their movement to the music with ease. The interaction is aimed to improve of the understanding of the ASD subjects on the correlation of their action to the tempo of WAS-5. We have assessed the effectiveness of the tempo sensing algorithm through preliminary experiments. This research aims to improve interactive social skills of ASD subject through the WAS-5 humanoid musical robot. The evaluation of the system shows promising tempo sensing with more fine tuning. Which Source Code Plagiarism Detection Approach is More Humane?

Oscar Karnalim and Lisan Sulistiani This paper contributes in developing source code plagiarism detection that is more aligned with human perspective. Three evaluation mechanisms that directly link human perspective with evaluated plagiarism detection approaches are proposed: think-aloud, aspect-oriented, and empirical mechanism. Using those mechanisms, a comparative study toward attribute- and structure-based plagiarism detection approach (i.e., two popular approach categories in source code plagiarism detection) is conducted. According to that study, structure-based approach is more effective than the attribute-based one; its signature aspect and resulted similarity degrees are more related to human preferences. In addition, such approach is related to most human-oriented aspects for suspecting source code plagiarism. Logic Error Detection Algorithm for Novice Programmers based on Structure Pattern and Error Degree

Yuto Yoshizawa and Yutaka Watanobe In recent years, the importance of programming skill is increasing due to advances in information and communication technology. However, the difficulty of learning programming is a major problem for novices. So we propose a logic error detection algorithm based on structure pattern and error degree. Structure pattern is an index of similarity based on abstract syntax trees and error degree is to measure appropriateness for feedback. In this paper, definition of the structure pattern and the error degree as well as the proposed algorithm method, are presented. Implementation and experiment with actual data are also considered.

38

Twitter and Online News analytics for Enhancing Post-Natural Disaster Management Activities

Kuhaneswaran Banujan, Banage T. G. S. Kumara and Incheon Paik A natural disaster is a natural event which can cause damage to both lives and properties. The detection of natural disasters is a significant and non-trivial problem. Social media (SM) is a powerful resource to improve the management of disaster situations. Post-disaster management can be improved to a great extent if we mine the SM properly because SM is capable of real-time nature of sharing the information. In this paper, we proposed an approach to enhance post-natural disaster management activities by identifying the correct location and disaster type. As the first step, we fetch the twitter posts using predefined keywords relating to the disaster from Twitter API. Those posts were cleaned and the noise was reduced at the second stage. Then in the third stage, we get the geolocation and disaster type. Named Entity Recognizer library and Google Maps Geocoding API was used for getting the geolocation. We did the same three stages for news which was fetched from News API. As a final stage, we compared the twitter datum with news datum to give the rating for the trueness of each Twitter post. “More accurate” rating was obtained for 24% of the posts. 15% and 13% of the posts showed “Moderately accurate” and “Less accurate” rating respectively. “No correlation” was obtained for 48% of the posts. The precision of 85% for Twitter posts filtering and 92% for News posts filtering were obtained when compared to the posts manually. We strongly believe that using this model we can alert the organizations to do their disaster management activities in a timely manner. We are planning to extend our work with the weather data and as well as with other social media to provide more scaled ratings. An Enhanced Hybrid MobileNet

Hong-Yen Chen and Chung-Yen Su Complicated and deep neural network models can achieve high accuracy for image recognition. However, they require a huge amount of computations and model parameters, which are not suitable for mobile and embedded devices. Therefore, MobileNet [1] was proposed, which can reduce the number of parameters and computational cost dramatically. The main idea of MobileNet is to use a depthwise separable convolution. Two hyper-parameters, a width multiplier and a resolution multiplier, are used to the trade-off between the accuracy and the latency. In this paper, we propose a new architecture to improve the MobileNet. Instead of using the resolution multiplier, we use a depth multiplier and combine with either Fractional Max Pooling [2] or the max pooling. Experimental results on CIFAR database show that the proposed architecture can reduce the amount of computational cost and increase the accuracy simultaneously. Classification of Online Judge Programmers based on Rule Extraction from Self Organizing Feature Map

Intisar Chowdhury and Yutaka Watanobe Computer programming is one of the most important and vital skill in the current generation. In order to encourage and enable programmers to practice and sharpen their skills, there exist many online judge programming platforms.Estimation of these programmers’ strength and progress has been an important research topic in educational data mining in order to provide adaptive educational contents and early prediction of ‘at risk’ learner. In this paper, we trained a Kohonen Self organizing feature map (KSOFM) neural network on programmers’ performance log data of Aizu Online Judge (AOJ) database. Propositional rules and knowledge was extracted from the U-matrix diagram of the trained network which partitioned AOJ programmers into three distinct clusters ie. ‘expert’, ‘intermediate’ and ‘at risk’. The proportional rules performed classification with an accuracy of 94% on a testing set. For

21 September (Fri) 10:55 - 12:15 (Room 521)

Session on Big Data

39

validation and comparison, three more predicting models were trained on the same dataset. Among them, feedforward multilayer neural network and decision tree have scored accuracy of 97% and 96% respectively. In contrast, the precision score for support vector machine was about 88%, but it scored the highest recall score of 99% in terms of identifying ‘at risk’ students. A Data Reconstruction Method for The Big-data Analysis

Masataka Mito, Kenya Murata, Daisuke Eguchi, Yuichiro Mori and Masahiko Toyonaga Recent years, the big-data becomes important for various business operations or sales strategies decisions. Contrarily several privacy issues prevent the advance of its analysis. There are some privacy-preserving methods, i.e. the anonymization or the extreme value records elimination or the fully encrypted analysis or so on. But the privacy cracking fears make it difficult to open the big-data outside researchers. We propose a big-data reconstruction method that uses intrinsically no the privacy data. The method uses only the statistical features of the big-data, i.e. its attribute histograms and their correlation coefficients. To verify whether some pieces of the valuable knowledge can be extracted from it or not, we evaluate the data by using SOM (Self-organizing-map) as one of the big-data analysis tools. The results show that the same bits of knowledge are extracted from our data and the big-data.

Learning Path Recommender System based on Recurrent Neural Network

Tomohiro Saito and Yutaka Watanobe Programming education has recently received broad attention due to demands for programming skills and information technologies. However, teaching materials and human resources for them remain a major challenge. So we propose a learning path recommendation system based on learner’s ability charts by means of a recurrent neural network. In briefly, learning path is constructed from submission history with a trial-and-error process of learners, and ability chart is a barometer of learner’s current knowledge. In this paper, an approach to construct learning path recommendation system by using ability charts and its implementation based on a sequential prediction model by a recurrent neural networks, are presented. Experimental evaluation with data of an e-learning system is also provided. A Study on Feature Extraction of Handwriting Data Using Kernel Method-Based Autoencoder

Quan Dang and Yan Pei We use kernel method-based autoencoder in feature extraction application and evaluate its performance with a public handwriting database. Neural network-based autoencoder is an unsupervised algorithm that tries to learn an approximation function so as to extract features from data. Kernel method- based autoencoder has the same function compared with neural network-based autoencoder, but uses kernel methods to implement linear and non-linear data transformation. We use a handwriting dataset to evaluate kernel-based autoencoder, and examine the result by mean square error estimator and structural similarity index for measuring image quality. We also investigate parameters of kernel functions to observe changes in the performance of the autoencoder. We found that performance of kernel method-based autoencoder depends on the selection of kernel function and its parameter. Analysis of Web Service Using Word Embedding by Deep Learning Takeyuki Miyagi, Incheon Paik and Rupasingha Arachchilage Hiruni Madhusha Rupasingha Service discovery is important issue when providing value-added services by composition. Existing approaches such as keyword or ontology matching have limitations within current

21 September (Fri) 10:55 - 11:55 (Room 524)

Session on Deep Learning

40

Web services because these approaches are working based on isolated services. To solve this problem, calculating service relationship is needed. When we calculate it, 4 properties are usually considered, functional similarity, quality of service (QoS), association of invocation, and sociability. In our previous research, we could calculate functional similarity and QoS by ontology or global social service network ¥cite{chen2015constructing}. But association of invocation and sociability has not been calculated from real world. In this research, we calculate them by using word embedding. Word embedding can find the relationship between services. In this research, we experiment to calculate similarity of Web API methods as services. By regarding the method call sequence as the input of word embedding, we observe how the method is related to other method. Finally, experimental results show that which method is related to other methods.

41

5-SENSE 2018 Technical Program Session on Taste and Odor Sensing (9:30 - 11:00, September 20, Room #531)

09:30 - 10:00 Rui Yatabe Taste Sensor and Odor Sensor

10:00 - 10:30 Mitsuru Tanaka Analytical methods for evaluation of food compounds

10:30 - 11:00 Fumihiro Sassa Collecting of spatial distribution of on-ground odor sources and odor visualization

Session on Perceptual Psychology (11:15 - 12:35, September 20, Room #531)

11:15 - 11:40 Emi Hasuo Time-shrinking illusion in the tactile modality: Comparison with auditory and visual modalities

11:40 - 12:05 Takahiro Kawabe Implementing visual illusions in the real world

12:05 - 12:35 Willy Wong On the Origins of the Power Function Exponent

Poster Session (13:50 - 15:30, September 20, Room #531 and foyer)

#1 The application of 360-degree video as a tool for standardized language testing material

#2 An acoustic analysys of preprosition phrases in English

#3 Psychological examinations of visual features of the heel professional wrestlers

#4 Nine different surface qualities can induce vection differently

#5 The perception of auditory icons by Japanese and Indonesian drivers – a cross-cultural study

#6 Development of color-temperature association in individuals with different color vision types

#7 Attention and impression toward complex images among individuals with different color vision types

#8 A study on the effect of sound on memorisation of symbolic images

#9 A report of a positive correlation between state anxiety and vection strength

#10 Pause duration influences impressions of speech style in English public speaking

#11 Inhibition of vection by grasping an object

#12 Identification of individuals based on the spatial arrangement of facial parts

#13 Effect of background luminance on the peripheral flicker illusion

#14 Effect of ISI on the vection-latency reduction induced by preceding vection

#15 Visual saltation illusion induced by flickering subjective figures

#16 Which spectral-change factors are indispensable to make Japanese speech intelligible? #17 Korean and Japanese speakers use different acoustic cues for hearing

Korean consonants: Measuring acoustic features of Korean lenis stops and Japanese voiceless and voiced stops

#18 Vection Can Be Modified by the Viewing Attitude of the Observers

#19 Rehabilitation Contents That Can Induce Stronger Vection are perceived more attractive

Session on Taste and Odor Sensing

Session on Perceptual Psychology

Poster Session

42

5-SENSE 2018 Poster Abstracts #1 The application of 360-degree video as a tool for standardized language testing material Laura Blanco1, Miharu Fuyuno, Gerard B. Remijn, Kazuo Ueda, Mikako Tomotari and Yoshitaka Nakajima.

1 Graduate School of Design, Kyushu University blanco.laura29@gmail.com

Since the 1960s, researchers in the field of CALL (Computer Assisted Language Learning) have studied new ways of using technology as a tool for language learning. In recent years, innovative technology such as immersive Virtual Reality has become widely available. Immersive VR enables users to interact with a virtual environment or a 360-degree video. The illusion of “presence” provoked in some cases by this visual and auditory immersion is still studied today. Applications using immersive VR have been developed in various fields such as medical, entertainment and cultural heritage education. However, its application in the field of language learning has not been fully investigated. Previous research pointed towards significant differences in listening comprehension test results when applying audio-only or audio and video stimuli. Following this tendency, we aim to investigate whether immersion has an influence in listening comprehension tests for Japanese university students of English as a Foreign Language. In order to examine the application of immersive VR technology, an experiment has been planned. The objective of this experiment is to study the efficacy of tools such as Head Mounted Displays (HMD) and 360-degree video for standardized language testing. A standardized listening comprehension test is re-created employing three different stimuli: audio-only (control), traditional video (on a flat screen) and immersive video (on a HMD). Listening comprehension conversation samples from a standardized test were recreated using a 360-degree camera and a microphone. The testing materials will be presented using a HMD and applied as a proficiency test for Japanese participants and the results will be analyzed in order to determine if the changes in perception affect the listening comprehension results of test-takers. #2 An acoustic analysys of preprosition phrases in English Xiaoyang Yu1, Yoshitaka Nakajima2, Yixin Zhang3, Takuya Kishida4 and Kazuo Ueda5

1 Graduate School of Design, Kyushu University yuxiaoyangat123@gmail.com

2 Dept. of Human Sciences/Research Center for Applied Perceptual Science, Kyushu University nakajima@design.kyushu-u.ac.jp

3 Graduate School of Design, Kyushu University zhangyixin0210@yahoo.com

4 Dept. of Human Science, Kyushu University kishida.takuya0119@gmail.com

5 Dept. Human Sciences/Research Center for Applied Perceptual Science, Kyushu University ueda@design.kyushu-u.ac.jp

The present study was on how spectral-change factors behave in English preposition phrases consisting of prepositions and noun phrases. Spectral changes of spoken English sentences in our new database were subjected to origin-shifted factor analysis. Three factors as in our previous study [Kishida et al. (2016). Front. Psychol. 7:517] appeared. One of the factors, which called high-frequency factor was closely related to a frequency range above 3300 Hz, and the factor scores were higher in the noun phrases than in prepositions. However, this phenomenon was not very clear in the others factors. Frequency components above 3300 Hz may play important roles to clarify noun phrases perceptually.

43

#3 Psychological examinations of visual features of the heel professional wrestlers Tomohiko Akagi1, Satoshi Ikehata2 and Takeharu Seno3

1,3Kyushu University, 2National Institute of Informatics Recently, professional wrestling is becoming much more popular in Japan. In professional wrestling, there are two types of wrestlers, i.e. baby face and heel wrestlers. We examined what visual characteristics are important for facilitating the attractiveness of the heel professional wrestlers. We conducted subjective estimations of the impressions of 30 Japanese pro-wrestlers and also did phenomenological approach to visual features of heel characters. The obtained subjective total attractiveness of the wrestlers, was significantly correlated with the attractiveness of the appearance and the face of them. The subjective fearfulness was not correlated with the total attractiveness. The image statistics showed us that the less colored and lower luminance can facilitate the impression of fearfulness. These results would suggest the visual features of the heel characters. #4 Nine different surface qualities can induce vection differently Hirotaro Sato1, Yuki Morimoto2, Chihiro Hiramatsu3 and Takeharu Seno4

1,2,3,4Kyushu University A new method for modulating vection strength will be reported. Vection can be modulated by the differences of the simulated surface materials used as vection stimuli. We used nine different surface materials, i.e. bark, fur, metal, ceramic, wood, leather, glass, fabric, stone. A tunnel made by these different surface qualities were used as vection stimuli. Forward self-motion was simulated. Three vection indices, i.e. latency, duration and magnitude of vection, were obtained as dependent variables. The results showed that vection can be stronger in bark condition and be weaker in ceramic, glass and metal conditions. We also modified the complexity of the tunnel shape, and obtained the same tendency. Vection strength can be affected by the surface qualities of the stimuli. #5 The perception of auditory icons by Japanese and Indonesian drivers – a cross-cultural study João P. Cabral1* and Gerard B. Remijn2

1 Graduate School of Design, Kyushu University 3ds17001e@s.kyushu-u.ac.jp

2 Dept. of Human Science/Research Center for Applied Perceptual Science, Kyushu University remijn@design.kyushu-u.ac.jp

*Corresponding author address: 4-9-1 Shiobaru, Minamiku Fukuoka, 815-8540, Japan Originally created for computer interfaces, auditory icons are one-shot sound messages that represent daily life events. After their introduction, auditory icons have been studied in several other fields, e.g., in the automotive and aviation industry, in medical fields, and the military. Regarding the automotive field, auditory icons have been used to inform events or situations about the vehicles’ condition, for example, the fuel level, or to prevent driver’s misbehaviors such as forgetting to lock the door, driving without fastening the seatbelts, or speeding. Regardless of the field and event they represent, the main objective of an auditory icon is to refer the listener to the information (the event) quickly and precisely. A review of the auditory icons’ literature revealed, however, that different sounds have been used as auditory icons to inform the same event. In addition, it is unknown whether auditory icons designed and tested with one user population, e.g., with listeners of one nationality, are interpreted similarly when presented to a different one. In this study, we investigated the perception and recognition of Japanese and Indonesian students of auditory icons representing vehicle driving and traffic events previously studied in the United States and Europe. Japanese and Indonesian participants listened to the same set of 15 auditory

44

icons and were asked to describe the sounds they heard using no more than two nouns and two adjectives. Before listening, both groups were informed that the icons concerned vehicle driving and traffic in general. Overall, the results showed the auditory icons with ubiquitous sounds (sirens, bells, voices) were interpreted similarly by most listeners of both nationalities. However, when describing auditory icons with the presence of high-frequency components, such as a ‘train crossing bell’, Japanese listeners described the icons according to the physical properties of the sounds using adjectives (e.g., ‘noisy’ and ‘high’), and/or the material characteristics by using nouns (e.g., ‘metal’) more frequently, compared to the Indonesian listeners. In addition, auditory icons with less well-known sounds were often correlated with events or objects in the listener’s environment, reflecting the listener’s cultural background and experience. #6 Development of color-temperature association in individuals with different color vision types Yuki Mori1 and Chihiro Hiramatsu2*

1 Graduate School of Design, Kyushu University 2 Department of Human Science, Faculty of Design, Kyushu University

chihiro@design.kyushu-u.ac.jp * Corresponding author address: 4-9-1 Shiobaru, Minamiku, Fukuoka, 815-8540, Japan

Many people tend to associate particular colors with physical temperatures. However, the mechanisms by which these associations occur are not well understood. They may be acquired though learning that occurs during development, and association patterns may differ among individuals with different color vision types. By investigating the color-temperature association patterns in individuals of different ages, with various color vision types, we aimed to reveal the factors that underlie these associations. Sets of two pieces of colored paper were presented to participants and they were asked to choose one colored paper that was felt warmer than the other. The combinations of colored paper were made from 20 Munsell colors. Regression analysis showed that hue and lightness of color had strong effects on the association between color and temperature in both dichromatic and trichromatic individuals, although the association patterns with lightness of color were not equal among participants. We also found that dichromatic adults showed similar association patterns to those of trichromats. Conversely, dichromatic children tended to show different association patterns from other participants. These results, which suggest a developmental/social learning aspect indicate that dichromatic people may be learning the same associations as trichromats (the majority in the population) during development. #7 Attention and impression toward complex images among individuals with different color vision types Chihiro Hiramatsu1*, Tatsuhiko Takashima2, Hiroaki Sakaguchi2, Satohiro Tajima3,4 and Takeharu Seno5

1 Department of Human Science, Faculty of Design, Kyushu University chihiro@design.kyushu-u.ac.jp

2 School of Design, Department of Visual Communication Design, Kyushu University 3 Department of Basic Neurosciences, University of Geneva

4JST Sakigake / PRESTO 5 Department of Human Science, Faculty of Design, Kyushu University

*Corresponding author address: 4-9-1 Shiobaru, Minamiku, Fukuoka, 815-8540, Japan Variation in color vision attributable to polymorphisms of L/M opsin genes is ubiquitous in human populations. Perceptual differences in color discrimination between dichromatic and trichromatic vision have been well studied. However, little is known about how differences in color vision affect vision beyond perception when viewing complex images. This study aimed to clarify how genetic polymorphisms at a few loci can influence individual differences in various aspects of vision such as attention to complex images and aesthetic impressions. Participants with either dichromatic

45

vision or trichromatic vision were asked to observe images of artistic paintings and answer questions regarding their impressions of the images. Gaze during art appreciation was tracked. Half of the participants with common trichromatic vision observed simulated dichromatic images. Gaze patterns in the first five seconds were associated with color vision types for images with certain color configurations, indicating that color vision differences affect bottom-up attention. The type of color vision also had a significant impact on the impression of colorfulness. Comparison of the results between innate dichromats and trichromats who observed simulated dichromatic images suggested that color impression was greatly influenced by the participants’ long-term experiences in their own color spaces. #8 A study on the effect of sound on memorisation of symbolic images Natalia Postnova1 and Gerard B. Remijn2

1Graduate School of Design, Kyushu University nataliapostnova@gmail.com

2Dept. of Human Science / Research Center for Applied Perceptual Science, Kyushu University remijn@design.kyushu-u.ac.jp

The effect of sound on the memorisation of visual symbolic images was examined using a simple memory task. Two experiments were conducted, testing the participants’ performance of the task in different sound conditions. Each participant was asked to memorise sequences of black-and-white symbols of a simple design, and then to recognise the symbols from the sequence using a program interface. During the presentation of the sequence, some of the symbols were accompanied by a short pure-tone sound, while others were presented in silence. There were several different stimulus conditions in which the sequences were presented: a no sound condition (none of the symbols in the sequence were presented with sound), a 100% sound condition (all of the symbols in the sequence were presented with sound), and a 25%, a 50% and a 75% condition, where the corresponding percentages of the symbols were presented with the sound and others without. For all the conditions the same short pure tone was used. Experiment 1 showed a positive effect of sound on symbol memorisation and recognition in a task where participants received a relatively basic combination of the stimulus conditions - no sound, 100% sound, and 50% sound conditions. Interestingly, the effect disappeared when the variation of conditions widened, that is, when the 25% and the 75% sound conditions were added to the three basic conditions. These results suggest a task-related effect, rather than a purely perceptual effect of the sound on the memorisation of visual symbolic images. #9 A report of a positive correlation between state anxiety and vection strength Ryosuke Ioka1, Takeharu Seno1 and Machia Okubo2

1 Kyushu University, 2 Senshu University Visually induced self-motion perception is named "vection". Vection strength can be modulated by various factors. Recently, a new method of modulating vection was reported by Seno et al. (2016). Vection can be facilitated by holding a Full glass of water while the participants observing vection stimulus. They named this method Full glass of water method. Then we hypothesized that the modulation of the state anxiety might be the hidden and most important reason for this vection facilitation. In this study, we measured the state anxiety by using visual analog scale before and after observation of vection stimulus. Vection was induced by a simple dot-optic flow. Three vection indices, i.e. latency, duration and magnitude of vection were obtained. The results clearly showed that the state anxiety and three vection indices, i.e. vection strength, were highly correlated. Our hypothesis was supported by the results. When the state anxiety is enhanced, perceived vection should be also enhanced.

46

#10 Pause duration influences impressions of speech style in English public speaking Shimeng Liu1, Yoshitaka Nakajima2, and Mark A. Elliott3

1 Graduate School of Design, Kyushu University liushimengdl@163.com

2 Dept. of Human Sciences/Research Center for Applied Perceptual Science, Kyushu University nakajima@design.kyushu-u.ac.jp

3 School of Psychology, National University of Ireland Galway, Galway, Ireland mark.elliott@nuigalway.ie

In the present study we aim to find the most suitable pause duration in public speaking and to establish a more efficient, objective index for the training of public speaking. A listening experiment was carried out varying pause duration systematically, in which, both comma and period pauses were changed with the same duration (of 0.1, 0.2, 0.4, 0.8, 1.6, and 3.2 s in each stimulus). Fifteen students (10 male and 5 female) from Kyushu University Undergraduate and Graduate Schools were invited to join the experiment as participants. The method and analysis were based on Uchida (2005, The Japanese Journal of Educational Psychology). Amongst 20 evaluation items, principal-component analysis allowed 5 components to be extracted, and a varimax rotation led to two factors that were useful for judging the quality of speech. These were “naturalness of speech” and “speech rate”.

Fig. 1. The average factor scores regarding naturalness of speech and speech rate.

The first extracted factor comprised a component reflecting the "naturalness of speech". The items that comprised this scale were "naturalistic", "easy to understand", and "clear". The second factor concerned "speech rate". The items that comprised this scale were "fast-talking", "hasty", "long pause duration (negatively)". As a result and regarding of the naturalness of speech, speeches with pause durations from 0.8 to 1.6 s received the most positive evaluations. On the scale: “speech rate”, scores became lower when pause duration became longer. We anticipate our results will be of use in training speakers in ‘when’ and ‘how long’ to pause, and will allow speakers to become more conscious and confident in controlling the timing and rhythm of their speech. Our study offers strong scientific support for the fact that ideal speakers share the same properties: When giving a speech, ideal speakers leave a pause of suitable pause duration to help the audience feel the naturalness of their speech; meanwhile this also makes the speech rate suitable. Our results will be discussed alongside a second study in which we collected data from English native-speakers, with whom we sought to establish whether there are differences between the two groups. #11 Inhibition of vection by grasping an object Masaki Mori1* and Takeharu Seno2

1 Graduate School of Media and Governance, Keio University Japan Society for the Promotion of Science, Japan.

morima@sfc.keio.ac.jp 2 Faculty of Design, Kyusyu University, Japan.

senosann@gmail.com *Corresponding author address: 5322 Endo, Fujisawa, Kanagawa 252-0882, Japan.

The present study investigated whether vection could be modified by an object grasping movement.

47

Twenty-five university students were asked to do one of the following four types of left hand movements while they were viewing a radial optic flow: (1) Grasping the hand-gripper strongly; (2) Holding the hand-gripper; (3) Clenching fist strongly; and (4) Open hand without having anything in their left hands (normal hand condition). The participants’ tasks were to keep pressing a button with their right hands while they were perceiving vection. Afterward each trial, they estimated the subjective strength of vection on a 101-point scale. The result showed that the vection was inhibited by strongly grasping the hand-gripper task more than by the other hand movements. Vection could be weakened by the object grasping movement. It might be suggested that vection could be inhibited by the presence of an object being grasped and also by the grasping movement itself. We speculated that the mechanism underlying this inhibition might be related to cognitive pressure, attentional load, power and muscle tonus, and multisensory and proprioception interactions. #12 Identification of individuals based on the spatial arrangement of facial parts Kana Uozumi1, Hiroyuki Ito2 and Masaki Ogawa3

1 Graduate School of Design, Kyushu University 2DS18092W@s.kyushu-u.ac.jp

2 Faculty of Design, Kyushu University 3 Faculty of Engineering, Mie University

This study compared the importance of the shapes of facial features with that of the spatial arrangement of facial features in facial identification of individuals. Before conducting the experiment, an averaged face was produced from faces of female university students using a morphing software. Three types of female facial stimuli were used, as follows: 1) original face, 2) parts-original face (facial parts of the averaged face were exchanged with those of an individual’s original face), and 3) arrangement-original face (facial parts of the individual’s original face were exchanged with those of the averaged face). Nine individuals’ faces were used as original faces, and subjects did not know any of them. Subjects performed face identification trials by key pressing after a face memorizing phase (30 seconds). The percentage of correct responses in the arrangement-original condition was higher than that in the parts-original condition. Reaction time (RT) for the arrangement-original faces was shorter than the RT for the parts-original faces. Therefore, the spatial arrangement of facial parts may be more important than the shapes of facial parts in the identification of individuals. However, it is possible that the facial outlines of the stimuli had some effect on facial identification. Hence, in a future study, we would like to conduct a similar experiment using facial stimuli with a masked outline. #13 Effect of background luminance on the peripheral flicker illusion Meidi Niikawa1 and Hiroyuki Ito2

1 Graduate School of Design, Kyushu University meidi.ann@me.com

2 Faculty of Design, Kyushu University The peripheral flicker illusion (Ito & Koizumi, 2017, i-Perception) is a phenomenon where blue or green objects are seen to flash twice or flicker when they suddenly appear on a red background. The flicker impression does not arise in central vision and for red objects. Thus, the authors suggest that the effect is related to interactions between L-cone and rod (or S-cone) activities. Here, we investigated the role of the red background in more detail, and varied the luminance of the blue and green objects, as well as that of the red background. Subjects evaluated the flicker impression, comparing that produced by physical flicker observed in central vision. When they felt no flicker impression, they assigned a score of 0, and when they felt the same amount of flicker impression as that caused by the physical flicker, they assigned a score of 10. We found that the evaluated flicker impression changed as the object luminance increased, resulting in an inverted U curve. The best object luminance, at which the flicker impression was the strongest, increased with the increase of the background luminance. However, the best object luminance never exceeded its background luminance. This suggests that the strength of the flicker impression was determined

48

by the ratio between the L-cone and rod (or S-cone) activities. #14 Effect of ISI on the vection-latency reduction induced by preceding vection Jing Ni1, Hiroyuki Ito2 and Masaki Ogawa3

1 Graduate School of Design, Kyushu University njliuxue@hotmail.com

2 Faculty of Design, Kyushu University 3 Faculty of Engineering, Mie University

Vection is an illusion of self-motion induced by visual motion. Because the illusion arises for stationary observers, visual and vestibular information contradict each other when vection arises. Thus, in a laboratory, several seconds are usually needed to start feeling self-motion. This is called latency. It is assumed that vection arises when visual information is more reliable than the vestibular information. This dominance of vision ceases when visual motion disappears. The purpose of this study was to measure this process. We assumed that feeling self-motion indicates visual dominance for a stationary observer and, therefore, after vection occurs, new vection in another direction could be induced with a shorter latency. After the first vection stimulus disappears, we introduced a time interval until the second vection stimulus appears (interstimulus interval, ISI). We hypothesized that when the ISI is short, the latency for new vection should also be short. Conversely, when the ISI is long enough, the latency should become equal to that in a control condition. As a first stimulus, we presented static random dots (control condition) or a vection-inducing stimulus moving vertically (vection-inducing stimulus condition). These were followed by an ISI of 0 - 4000 ms before presentation of the second stimulus. The second stimulus was a vection-inducing stimulus that moved horizontally. We measured the vection latency for the second stimulus. We found that the vection-inducing stimulus condition elicited a shorter vection latency for a second vection-inducing stimulus than did the control condition. This difference decreased as the ISI increased. These results indicate that while pre-activation of the visual motion processing system by the first vection-inducing stimulus contributed to visual dominance over the vestibular system at the presentation of the second stimulus, the dominance disappeared when the pre-activation ceased due to a longer ISI. #15 Visual saltation illusion induced by flickering subjective figures Hiroyuki Ito

Faculty of Design, Kyushu University ito@design.kyushu-u.ac.jp

After a visual object flashes twice at the same position (first and second flashes), followed by another flash (third flash) at a displaced position, the visual saltation illusion arises, whereby the second flash is perceived to occur at a position between the first and third flashes (Geldard, 1976). This study investigated whether subjective figures could induce the visual saltation illusion. The Ehrenstein figure and Kanizsa triangle were used as subjective figures. In addition, a texture-defined figure (a square filled with horizontal stripes presented on a background filled with vertical stripes) was tested. As a control, luminance-defined figures (a disk, a triangle, or a square in black and presented on a white background) were also tested. The duration of flashes was 50 ms and the interstimulus intervals were 100 ms. Retinal eccentricity of the stimuli was also varied. Subjects evaluated the illusion using a 6-point scale (0-5). We found that 1) the saltation illusion did not occur in near central vision, while the illusion was strong in far peripheral vision, 2) the strength of the illusions produced by a black disk and the Ehrenstein disk was nearly equal, and 3) the illusions in the Kanizsa triangle and the texture-defined square conditions were stronger than the black triangle and the black square, respectively. This suggests that the subjective figures are favored by the illusion rather than the luminance-defined figures. The illusion may not need luminance motion signals. The weaker position cues of the subjective figures may enhance the illusory shift of the second flash.

49

#16 Which spectral-change factors are indispensable to make Japanese speech intelligible? Takuya Kishida1, Yoshitaka Nakajima2, Kazuo Ueda3, Gerard B. Remijn4 and Seiya Umemoto5

1 Dept. of Human Science, Kyushu University kishida.takuya0119@gmail.com

2 Dept. of Human Science/Research Center for Applied Perceptual Science, Kyushu University nakajima@design.kyushu-u.ac.jp

3 Dept. of Human Science/Research Center for Applied Perceptual Science, Kyushu University ueda@design.kyushu-u.ac.jp

4 Dept. of Human Science/Research Center for Applied Perceptual Science, Kyushu University remijn@design.kyushu-u.ac.jp

5 21st Century Program, Kyushu University 1NC15002K@s.kyushu-u.ac.jp

A listening experiment was conducted to clarify essential cues to perceive speech. We applied a modified version of principal component analysis to sum up spectral change of Japanese speech sounds into small number of factors. The first 2, 3, and 4 principal components were rotated with varimax rotation yielding three sets of spectral-change factors. The Japanese speech was resynthesized as noise-vocoded speech. Spectral change of the noise-vocoded speech was reconstructed from 2, 3, and 4 factors selected from each available set of the factors. Sixteen native speakers of Japanese listened to the noise-vocoded Japanese speech and reported morae (syllable-like phonological units) which they heard. The number of factors affected intelligibility: Mora identification performances were around 8-30% in the 2-factor conditions, around 37-60% in the 3-factor conditions, and over 80% in the 4-factor condition. Among the 3-factor conditions, the lowest mora identification was observed when the factor located around 1100 Hz was omitted from the resynthesis. Nakajima et al. (2017, Sci. Rep.) argued that the factor corresponds to what is called ‘sonority’ in phonology. The results indicate that perceiving sonority and enough spectral cues are both required for intelligible speech perception. #17 Korean and Japanese speakers use different acoustic cues for hearing Korean consonants: Measuring acoustic features of Korean lenis stops and Japanese voiceless and voiced stops Yubin Sung1* and Hiroyuki Mitsudo2

1 Graduate School of Human-Environment Studies, Kyushu University yubiso91@gmail.com

2 Graduate School of Human-Environment Studies, Kyushu University hmitsudo@lit.kyushu-u.ac.jp

*Corresponding author address: 4-9-1 Shiobaru, Minamiku, Fukuoka, 815-8540, Japan

The Korean lenis stop (ㄱ, ㄷ, ㅂ), a category of Korean consonants, is generally pronounced as a voiceless sound (/k/, /t/, /p/), but as a voiced sound (/g/, /d/, /b/) when located between vowels. Thus, Korean speakers are expected to match the Korean lenis stop in the initial position of a word to the Japanese voiceless stop. However, our previous study found that Korean speakers matched the phoneme of lenis stops to the phoneme of Japanese voiced stops (Sung and Mitsudo, 2018, the annual meeting of the Japanese Society of Cognitive Psychology). These results suggest that Korean speakers acoustically match the lenis stop in the initial position to the Japanese voiced stop. Our results also showed that Japanese speakers matched lenis stops to voiceless stops. These results imply that Korean and Japanese speakers hear the lenis stop using different acoustic cues. In the present study, to investigate what acoustic cues produced the results of Sung and Mitsudo (2018), we analyzed some acoustic features of the speech sounds used in Sung and Mitsudo (2018) by using praat. In particular, we measured the VOT (voice onset time) of the lenis stops pronounced by native Korean speakers and the Japanese voiceless and voiced stops pronounced by native Japanese speakers. Additionally, we measured the F0 (fundamental

50

frequency) of the Japanese voiceless and voiced stops pronounced by native Japanese speakers. Consequently, the VOT of the native Korean speakers’ (N = 4) lenis stops tended to be more similar to that of the native Japanese speakers’ (N = 4) Japanese voiceless stops than that of Japanese voiced stops. Furthermore, the F0 of the Japanese voiced stops tended to be lower than that of Japanese voiceless stops. These results are generally consistent with previous studies that investigated acoustic features of Korean and Japanese stops. The present study suggests that when matching lenis stops to Japanese voiceless or voiced stops, Korean speakers use F0 rather than VOT, whereas Japanese speakers use VOT. Taken together, Korean and Japanese speakers may use different acoustic cues for hearing lenis stops.

#18

Vection Can Be Modified by the Viewing Attitude of the Observers Hirotaro Sato1, Takeharu Seno2 and Gerard B. Remijn2

1 School of Design, Kyushu University, Japan 2 Faculty of Design, Kyushu University, Japan

Visually-induced self-motion perception, or “vection”, was investigated in participants who received three types of experimental instruction while watching vection-inducing stimuli consisting of expanding dot motion. In the Vection-Positive condition, participants were asked to be sensitive to vection perception and try to actively perceive vection. In the Vection-Negative condition, they were asked to resist vection perception, while in the Control condition, they were asked not to have any viewing attitude toward the vection stimulus. Each instruction condition was repeated five times, and the results showed that the type of instruction highly modified vection strength. Vection-Positive and Vection-Negative instructions facilitated and inhibited vection, respectively, while to the vection strength induced by the Control instruction was in between these two. In future vection research, this high sensitivity to experimental instruction and the viewing attitude of the participants to vection stimuli should be considered fully in the experimental design and interpretation of results.

#19

Rehabilitation Contents That Can Induce Stronger Vection are perceived more attractive Ryosuke Ioka1, Rika Tanka2, Takeharu Seno3 and Hiroyuki Matsuguma4

1,2 Graduate School of Design, Kyushu University, Japan 3,4 Faculty of Design, Kyushu University, Japan

Vection (visually induced illusory self-motion perception) can enhance the attractiveness of the contents for the rehabilitations. Some people experienced traffic accidents or brain damages should do some severe rehabilitations for getting better conditions. However, the rehabilitations are sometimes no attractive and hard for them, especially for the elderly people. Thus, to make the contents of the rehabilitation more attractive is an urgent and important topic in this field. To present vection stimuli as these rehabilitation contents, can improve the attractiveness of them, we thought. Thus, we examined whether the visual contents that could induce stronger vection, could be perceived more attractive as the rehabilitation contents. We prepared four different movies of meaningful vection scenes and one control meaningless vection movie. These four were, forward-moving expanding optic flows in the cave, on the river, in the sky and in the space. As a control vection movie, we presented a simple dot-optic flow. We obtained two vection indices, i.e. latency, and subjective strength of vection and also obtained the subjective attractiveness of these contents by using visual analog scales and also by the button pressing by the participants. The results clearly showed that the perceived attractiveness of these contents and the perceived vection strength were highly positively correlated. When the contents induced stronger vection, those contents were perceived more attractive. We believe that by enhancing vection perception, the contents can be much more attractive. We speculated that by making the contents more attractive, the elderly people would do rehabilitation more often and keep it in much longer time.

51