日時: 12月7日(水)3限(13:30-15:00)

会場: L1

司会: 川上 朋也
吉田 貴信 1651123: M, 1回目発表 ソフトウェア設計学 飯田 元
title: Paper introduction: "Interactive Recovery of Requirements Traceability Links Using User Feedback and Configuration Management Logs"
abstract: Traceability in software development refers to the ability to trace the relationships between software products (e.g. requirement specification, design specification, source code file). It has been reported that exploiting traceability can effectively reduce the time required for software maintenance. Furthermore, for achieving traceability in software development, it is required to manually determine and construct the relationships between software products. Such relationship, or called as Traceability Link (Link), can be done by human-force or system's using specified method. In addition, during the process of linking, low left-outs as well as small error rate are demanded for improving the traceability. Previous linking system utilizes similarity-based method to construct links between software products. However, the left-outs of links and the error rate achieved by such system are around 50%, which is considered to be not practical. Therefore, this presented paper proposes a linking system which performs linking based on"The modification logs of source codes" and "The link validations by users and their feedbacks", The authors also performed an experiment for evaluating the system's efficiency. As a result, compared to the traditional method, the left-outs of links as well as the incorrect linking rate are decreased by 20% and 5% respectively. In this presentation, based on what we learnt from the introduced paper, we will discuss how it relates to the design process and the quality of design specification in software development. As well, we will mention about our future research plan.
language of the presentation: Japanese
藤原 聖司 1651093: M, 1回目発表 ユビキタスコンピューティングシステム 安本 慶一
title: Activity Recognition System in Smart Home using Analog-PIR-Sensor
abstract: Recently, smart home applications such as home appliance control and adult care service are expected. In order to realize these applications, it’s important to recognize the various human living activities accurately and cheaply. There are many activity recognition approaches for smart home, such as the techniques using door opening / closing sensors, PIR sensors, and power consumption sensors. However, these studies have the problems that we cannot recognize between the similar tiny activities at the same place. In order to solve this problem, we propose analog-output-PIR-sensor-based activity recognition technique which recognize the detailed activities of the user. We use the machine learning to recognizing the activities of the user. we use output patterns and frequency components of the analog-output-PIR-sensors as feature values. In order to evaluate our proposed system, we performed the experiment that three subjects performed four activities with sitting on a sofa. As a result, we achieved F-Measure: 57.0%.
language of the presentation: Japanese
発表題目: アナログ出力焦電型赤外線センサを用いたスマートホーム内行動認識
発表概要: 近年のユビキタス情報技術の発展に伴い,省エネ家電制御や高齢者見守りシステムなど生活を支援するサービスへの応用が期待されている. これらのサービスを実現するには,居住者の多様な屋内生活行動を高精度かつ安価に認識できることが重要である. スマートホーム内向け行動認識手法は多数提案されており,これまでにドア開閉センサや焦電型赤外線センサ,消費電力センサを利用したスマートホーム内における生活行動認識手法が開発されている. しかし,これらの先行研究では,居住者の位置とその行動が1対1で対応しない位置非依存行動を高精度で認識できないという問題があった. そこで,本研究ではアナログ出力焦電型赤外線センサを開発し,人の身振りの大きさなどユーザの動作の違いを検出することで,この問題を解決する.提案手法では,人の動作の違いによりアナログ赤外線センサの 出力パターンや周波数成分に違いが生じることに着目し,機械学習を用いて位置非依存行動を認識する. 提案手法の有効性を評価するため,スマートホーム設備(1LDK)にて計3名の被験者にソファに着座した状態で4種類の異なる行動を行う実験を実施した. その結果,平均F値が57.0%でユーザの行動を認識可能であることを確認した.
日高 真人 1651090: M, 1回目発表 ユビキタスコンピューティングシステム 安本 慶一
title: A study on real-time tourism planning system based on curation of CGM
abstract:In recent years, along with an increase in foreign tourists visiting Japan, improvement of the environment that tourists can enjoy sightseeing comfortably is demanded. To realize comfortable sightseeing, it is indispensable to enhance tourist information contents. However, it is difficult to make a sightseeing plan that suits you from a lot of static tourist information. In this research, we aim to create a system to compute the tour plan in real time according to individual needs from the photograph (CGM: Consumer Generated Media) of tourist spots obtained by participatory sensing. In this presentation we report experiments verifying the effectiveness of the method of recommending recommendation spot using curation in the way from spot A to spot B.
language of the presentation:Japanese
発表概要:近年、訪日外国人旅行者の増加に伴い、観光客が快適に観光を満喫できる環境整備が求められている。快適な観光の実現には、観光情報コンテンツの充実か必要不可欠てある。しかし、多数の静的な観光情報から自分に合った観光プランを作成するのは難しい。 本研究では、参加型センシングで得られた観光地の写真(CGM:Consumer Generated Media)から、個人のニーズに合わせて観光プランを実時間編纂(キュレーション)するシステムの作成を目標とする。本発表ではスホットAからスホットBまての道中にあるお勧めスホットをキュレーションを用いて推薦する手法の有効性を検証した実験について報告する。
則永 悠 1651088: M, 1回目発表 知能システム制御 杉本 謙二
title:Motion intention estimation using EMG and environmental information abstract: Electromyogram (EMG) signals are often used in myoelectric prothetic and robot control. However, current myoelectric systems can not perform dexterous motions like human because it is difficult to accurately estimate the detailed motion intention from the EMG signal. Therefore, in this research, we attempt to achieve high accuracy of motion intention estimation from EMG signals with environmental information acquired by visual sensor etc. In this presentation, we describe the related studies and their remaining issues. Then we present our research plan and experimental platform. langage of the presentation : Japanese 発表題目:筋電位と環境情報を利用した動作意図推定 発表概要:筋電義手の操作には、しばしば筋電位信号が利用される。 しかし、筋電位信号から詳細な動作意図を正確に推定することは難しいため、 現在の筋電義手では人の手のような精密な動作が行えない。 そこで本研究では、筋電位信号に加えて視覚センサなどで 取得される環境情報を利用することで動作意図推定の高精度化を検討する。 本発表では、先行研究や残された課題について言及し、問題解決の着想、構築した実験環境や研究計画について述べる。
LIU XINRAN 1651129: M, 1回目発表 自然言語処理学 松本 裕治
title: A Macro-structure Analysis of Research articles in EMNLP
abstract: The study presented in this paper examined the macro-structures (MSs) of research articles (RAs) in EMNLP. We attempted to unified MS model to represent the discipline under investigation and provided explanations of intra-disciplinary variability, this study set out to examine how MSs of RAs in EMNLP may formed and present a linguistic description of the schematic organization. 802 articles of EMNLP were collected from 2014-2016. Their main sections were subjected to a series of analyses. Results show that some regular patterns of structural organisation can be detected in the RAs belongs to a specific academic journal. Implications for knowledge discovery through structural document understanding and further applications like paper recommendation.
language of the presentation: English
野原 有実 1651087: M, 1回目発表 計算システムズ生物学 金谷 重彦
title:Analysis and control of motion of Brownian particle in an optical trap
abstract:The radiation pressure exerts on small particles when a focused laser beam is irradiated on them. Optical tweezers utilize the radiation pressure to capture and operate small particles under a microscope without contact and damage. As a problem during operation, a captured particle sometimes gets out of the trapping spot because of the Brownian motion of it. To solve this problem and to improve capture and operation accuracy, we analyze and control the motion of Brownian particle in an optical trap in this study. A previous work of analysis estimates particle size and 3D-position by fitting intensity data of hologram which records the information of phase distribution as a result of interference between the light scattered by a particle and a reference beam. In this talk, I will introduce a relevant study and make a report of my future research plan.
language of the presentation: Japanese

会場: L2

司会: 藤本 まなと
政野 博紀 1651097: M, 1回目発表 モバイルコンピューティング 伊藤 実

title: Study of Real‐Time Route Schedules for Demand Responsive Transport with Bus Transfers

abstract: Demand Responsive TransportDRThas been increasingly attractive for transportation in sparsely populated areas, where public transport services rarely exist. However, most of current DRT systems run with the help of a central operator, who manually schedules bus routes according to user queries, which is inconvenient and not efficient. Thus, more flexible DRT with automatic route scheduling is of great interest. In this research, we are going to propose a DRT system, which dynamically schedules bus routes. To reduce the DRT system cost (total bus running distance), navigation for user walking and bus transfer is also considered.

language of the presentation: Japanese

発表題目: オンデマンドバスの乗り換えを含むリアルタイムルートスケジューリングの検討

発表概要: 近年,特に過疎地域において,固定路線バスの不採算路線の代替手段としてデマンドバスが注目され,実用化されている.しかし,多くのデマンドバスが予約にオペレータの介在が必要であるなど利便性が高いとは言えず,また採算性も低い.これに対し,先行研究では自由な経路を持ち,より低遅延で予約の受理率が高いデマンドバスを自動でスケジューリングするシステムの研究が行われている.本研究では過疎地域においての運行も想定して,徒歩移動のナビゲートとデマンドバス間の乗り換えをナビゲートするデマンドバスのシステムを提案する.本システムにおいて,デマンドバスは自由な経路を走行するが,ユーザに対しデマンドバスの迂回が少なくなる乗降地点まで徒歩経路を案内し,乗り換えも案内する.これによりデマンドバスの迂回経路を削減し,運行の効率化を図る.

大内 勇磨 1651021: M, 1回目発表 インタラクティブメディア設計学 加藤 博一
title: 3D Ground Reaction Force Visualization for Sprint Training
abstract: In this study, we design a method for visualizing 3D ground forces for sprint training. We use mixed reality technology to simultaneously display 3D ground reaction forces and sprinter’s form in virtual space. By using this method, athletes can analyze 3D ground reaction forces and their form intuitively.
language of the presentation: Japanese
発表題目: スプリントトレーニングのための三次元床反力の提示方法の検討
発表概要: スプリント走時の床反力データと走行フォームを同時に確認可能な複合現実感型トレーニング支援システムを提案する.本システムでは,ビデオ映像から走者を切り出し,仮想空間内で床反力データと統合し可視化する.これにより,様々な角度からフォームおよび床反力データを同時に確認することを可能とする.
矢萩 拓夢 1651116: M, 1回目発表 ネットワークシステム学 岡田 実
title: Accurate Positon Detection using MUSIC Method in LCX-MIMO system.
abstract: Recently, Demand for performing position detection by terminal or tag by wireless on indoors and outdoors is increased. Installation of multiple access points is needed for covering a wide area in the method of using RFID or Wi-Fi. LCX (Leaky coaxial cable) is advantageous in forming an elongated area and we proposed that position detection using it. So far characterization considering line-of-sight propagation path was done but characterization considering fading spatial propagation path in was not done. In this research, we proposed that position detection suitable for multipath propagation.
language of the presentation: Japanese
発表題目: MUSIC法を用いたLCX-MIMOシステムにおける高精度端末位置検出
発表概要: 近年、屋内外においてワイヤレスで端末やタグ位置検出を行う需要が高まっている。RFID、Wi-Fiなどを用いる方式では広いエリアをカバーするために多数のアクセスポイントを設置する必要がある。LCX(漏洩同軸ケーブル)は細長いエリアの形成に有利であり、本研究ではこれを用いた位置検出方法を提案している。これまで、見通し伝搬路のみを考慮した特性評価が行われたが、空間伝搬路で起こるフェージングを考慮した評価を行われていない。本研究では多重路伝搬に適した位置検出法を提案する。
森 巧磨 1651110: M, 1回目発表 知能コミュニケーション 中村 哲
title: Distilled Bi-directional Recurrent Neural Network models for End-to-End Automatic Speech Recognition
abstract: Automatic speech recognition (ASR) systems transcribe human speech into words. They are the key component of many application, such as speech summarization, call center, speech translation. The most recent technique used in ASR is end-to-end based on deep learning. The model based on end-to-end method is simple that does not require any phonetic information due to the use of character levels. However, this is problematic because the model becomes larger and the amount of calculation increases, so that the response at the time of use becomes slower. In addition, we succeeded in producing better performance by learning a child model as usual, with a model trained in previous research as a teacher to train other child models, which is called distillation. The proposed method uses this as End-to-End ASR. Thus, while the recognition accuracy is slightly lowered, the response speed can be increased.
language of the presentation: Japanese
村瀬 行俊 1651108: M, 1回目発表 知能コミュニケーション 中村 哲
title: Feature Inference by Label Propagation on external knowledge for Dialog State Tracking
abstract: Dialog state tracker(DST) is researched as one of the most important module of spoken dialog system(SDS). DST analyzes users' intentions by their utterances with dialog histories. Topics talked in dialog are also important to be analyzed by knowledge that is like human. Wikidata, which is scripted knowledge on Web, is used as knowledge base and run label propagation on the knowledge base. We investigated that the accuracy of dialog state tracking is increase by using knowledge and inference method.
language of the presentation: Japanese
溝口 拓也 1651102: M, 1回目発表 光メディアインタフェース 向川 康博
title: Analysis on frequency response using ToF camera for visualizing inner structure of human body
abstract: In Kinect v2 which is one of the Time-of-Flight cameras, distance is measured by detecting phase lag in reflected light signals modulated with three different frequencies. The amplitudes of the three waves are different due to subsurface structure. This is due to the difference in frequency response of the internal structure of translucent objects due to subsurface scattering. Therefore, we made an investigation with the assumption that the amplitude of the three waves themselves reflects the internal structure of the human body. In this presentation, as a clue to visualize the internal structure of the human body, we describe the approach focusing on the amplitude difference of the three waves.
language of the presentation: Japanese
発表題目: 人体内部構造の可視化に向けたToFカメラに対する周波数応答の解析
発表概要: Time-of-Flightカメラの一つであるKinect v2では,3つの異なる周波数で変調された光の反射波における位相遅れから距離を計測している。 その3つの波の振幅は,表面下の構造によって異なる. これは,表面下散乱により半透明物体の内部構造の周波数応答に差が生じることが原因である. そこで我々は3つの波の振幅自体が,人体の内部構造を反映するという仮定を置き,調査を行った. 本発表では,人体の内部構造の可視化の為の手がかりとして,3つの波の振幅差に着目した取り組みについて述べる.

会場: L3

司会: 進藤 裕之
向井田 一平 1651107: M, 1回目発表 環境知能学 萩田 紀博
title: Survey about Continual Conversations of Communication Agents for Elderly
abstract: In an aging society, elderly people who live alone are increased. For elderly people, lack of communication between other people is a factor in depression and dementia, and is related to suicide and solitary death. Therefore, it is important to put isolated elderly people in more communicative. In this study, we focus on conversational agent for increasing the number of opportunities elderly people continually talk. We especially discuss two important methods, interest of utterance contents, impression of agent.
language of the presentation: Japanese
発表題目: 高齢者のための対話エージェントの継続的な対話に関する研究調査
発表概要: 現在,高齢社会を迎えた日本において,日常的に人とコミュニケーションをとる機会がなく社会的に孤立した高齢者が増加している. 高齢者にとって,コミュニケーション不足は鬱や認知症の要因となり,自殺や孤独死につながる. そのため,社会的に孤立した高齢者に,コミュニケーションの機会を増やすための対策が重要な課題となっている. そこで本研究では,音声対話エージェントを用いて,高齢者への継続的な発話機会の提供を目的とする. 本発表では特にエージェントとの継続的な対話に重要な,発話内容の新奇性,印象の良さについて述べる.
藤井 公大 1651092: M, 1回目発表 大規模システム管理 笠原 正治
title: Detection Scheme of Isolated Areas Based on Information about Evacuees' Movement
abstract: When a large-scale disaster occurs, it is essential to quickly grasp the post-disaster situations. For this purpose, most of the existing techniques use aircraft and/or satellite but they require initial waiting times due to flight plan and orbit period of satellite. We expect that the automatic evacuation guiding scheme, which is proposed by Komatsu et al., can be applied to the estimation of post-disaster situations. The automatic evacuation guiding scheme can automatically estimate blocked road segments in a real-time manner, with the implicit interactions between evacuees and their mobile nodes. In this presentation, we propose a scheme to detect isolated areas, in which evacuees cannot find out any route to safe places, using the information about blocked road segments. Through simulation results, we show the fundamental characteristics of the proposed scheme.
language of the presentation: Japanese
発表題目: 避難行動情報を用いた孤立領域検出方式に関する検討
発表概要: 大規模災害が発生した場合,災害発生後の状況を迅速に把握する必要がある.既存技術の大部分では,航空機や衛星の使用による状況把握を試みているが,飛行計画や衛星の軌道周期に制約があり,稼働するまでに時間を要する.小松らが提案した自動避難誘導方式は,災害発生後の状況の推定にも適用できると考えられる.自動避難誘導方式は,避難者と避難者が所持するモバイル端末の連携により,通行不能箇所をリアルタイムで自動的に推定することができる.本発表では,通行不能箇所の情報を用いて,避難者が避難所への経路を見つけることができない領域(孤立領域)を検出する手法を提案し,シミュレーション結果を用いて提案方式の基本特性を示す.
与那嶺 俊 1651125: M, 1回目発表 情報基盤システム学 藤川 和利
title: [paper introduction] ReFormat Automatic Reverse Engineering of Encrypted Messages
abstract: Malware continues to be one of the major threats to Internet security. Some malwares use encryption algorithms to hide their connections with attackers. ReFormat, a program-based system proposed by Wang et al., uses taint tracking analysis to detect encryption algorighm in malware. Taint tracking is one of the control flow analysis method based on dynamic binary instrumentation which enables to trace execution flow of malware. In this presentation, I introduce automatic binary analysis technique through previous work.
language of the presentation: Japanese
発表題目: 暗号化メッセージの自動解析手法
発表概要: マルウェアはインターネットセキュリティにおける脅威の1つである。マルウェアの中には暗号化を用いて攻撃者との通信を隠蔽するものがある。Wangらが提案したReformatはテイント解析を用いてマルウェアが持つ暗号化処理を検出する。テイント解析とは動的バイナリ軽装を用いた制御フロー解析手法の1つであり、マルウェアの実行フローの追跡を可能とする。本発表では、先行研究を通してバイナリ解析の自動化手法について紹介する。
山本 諭 1651119: M, 1回目発表 数理情報学 池田 和司☆
title: Neural decoding of subjective time for decision making
abstract: When we make a decision, it is important to detect the time to rewards. In the previous study, it is known that a time parameter like time discount rate can explain the behavioral properties. We have not considered the relation between time parameters and subjective time. We check this relation in behavioral and neural level. We conduct the neural decoding of subjective time while measuring neural activity by functional magnetic resonance imaging. By checking relations with a time parameters, we propose the performances of behavioral task and decodability of time measured by this method as indexes of personal attributes.
language of the presentation: Japanese
宮本 康平 1651106: M, 1回目発表 生体医用画像 佐藤 嘉伸

title: Analysis of fiber structure of skeletal muscles around the hip joint using CT images

abstract: Biomechanical simulation is effective for muscle kinetics analysis. Providing muscle fiber information to skeletal muscle model enable patient specific simulation. The muscle fiber is estimated by diffusion MRI. It is not use in clinical practice, so we cannot estimate patient specific fiber orientation using CT image. Therefore, in this project, we will estimate the orientation of skeletal muscle fibers from clinically available modalities such as CT image. In this presentation, I present a result of preliminary experiments and future work.

language of the presentation: Japanese

発表題目: CT画像を用いた股関節周辺骨格筋の線維走行解析

発表概要: 筋動態解析においてバイオメカニクスシミュレーションが有効である。骨格筋形状モデルに筋線維情報を与えることで、患者固有の詳細なシミュレーションが可能になる。これまで、Diffusion MRIを用いた筋線維構造の抽出が行われているが、一般的な診察では用いられていない。そこで本研究では臨床の現場で用いられるCT画像からの骨格筋線維構造の抽出を目的とする。今回の発表ではCT画像に比べ高解像度で撮影されたVisible Korean Human Datasetの光学画像を用いた股関節周辺の骨格筋線維方向の解析を行った結果と今後の課題について報告する。

村瀬 由基美 1651109: M, 1回目発表 ロボティクス 小笠原 司
title: Visual-based navigation using scene understanding in outdoor environment for autonomous mobile robot
abstract: Self-driving system and driver support system are expected to be used not only for automobiles but also for personal mobility, moving on pavement such as electric wheelchairs. In the previous studies, it is possible to localize in an outdoor environment where the appearance changes depending on weather and time by comparing sequences of images. However it is not assumed to move on the road. In addition, it is possible to detect the travelable area by dividing and understanding the image of the environment according to object classes: road, vegetation, building etc. Therefore, in this research, we aim to improve path planning and localization by using visual-based navigation by using scene understanding in a mobile robot that assumes to move in pavement in an environment that appearance changes. In this presentation, as a progress, we will show trial and consideration of driving in outdoor environment using a mobile robot and the results of application of mapping method based on images in indoor environment.
language of the presentation:Japanese
発表題目: 屋外自律移動ロボットのためのシーン理解を用いた視覚ベースナビゲーション
発表概要: 自動運転技術や運転支援技術は自動車だけでなく,電動車椅子のような歩道を走行するパーソナルモビリティへの利用も期待される. 先行研究では,連続画像の比較を行うことで天候や時間帯によって見え方が変化する屋外環境において自己位置推定を可能としているが,車道を走行する場合しか想定されていない. また,環境の画像を路面や植生,建物等の領域によって分割し理解することで走行可能領域の検出が可能となる. そこで,本研究では見え方が変化する環境において歩道走行を想定した移動ロボットで,シーン理解を用いた視覚ベースナビゲーションを行うことにより経路計画や自己位置推定精度の向上を目指す. 本発表では,進捗として移動ロボットを使った屋外走行の試行と考察,画像に基づいた地図生成手法の屋内環境への適用結果を示す.