コロキアムB発表

日時: 12月16日(木)3限(13:30~15:00)


会場: L2

司会: 藤本 雄一郎
横路 海斗 M, 1回目発表 サイバネティクス・リアリティ工学 清川 清, 向川 康博, 内山 英昭, 磯山 直也
title: Work Assistive System Removes Visual Noise
abstract: Desks where we work are often cluttered with various objects such as computers, documents, and small items over time. In a cluttered environment, visual noise, which is multiple visual stimuli unrelated to the work, is continuously introduced into the visual field, which has been shown to reduce the cognitive function of the brain and decrease concentration and information processing ability. In addition, it is thought that a cluttered desk makes it more time-consuming to find things and breaks concentration on work. As a result, cluttered environments and desks are not suitable places to work because they cause a decrease in concentration and work efficiency. However, in order to reduce the effects of a cluttered environment, it is necessary to keep the desk clean. It is difficult for some people to keep their desks in a clean condition. In this study, we propose a system that visually removes only the visual noise that interferes with concentration on work. By visually removing objects that are irrelevant to the work, we aim to improve concentration and work efficiency.
language of the presentation:Japanese
発表題目: 作業時の視覚的ノイズを除去する作業支援システム
発表概要: 作業を行う机には,時間とともに,PCや書類,小物など,さまざまな物が散らかることが多い.ものが散らかった環境では,視野に作業と無関係な複数の視覚的刺激である視覚的ノイズが入り続けることで,脳の認知機能が低下し,集中力や情報処理能力が下がることが分かっている.また,散らかった机では,物を探すのに時間がかかり,作業への集中が途切れてしまうことが考えられる.これらのことから,散らかった環境や机は集中力や作業効率の低下を引き起こすため,作業する場所としては向いていない.しかし,散らかった環境による影響を軽減するためには机の上をきれいにする必要がある.また,机の上をきれいな状態で維持するのは人によっては困難である.そこで本研究では,作業への集中を阻害する視覚的ノイズのみを視覚的に除去するシステムを提案する.作業と無関係な物体を視覚的に削除することで,集中力や作業効率の向上を目指す.
 
石丸 大稀 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, 清川 清, 諏訪 博彦, 藤本 まなと
title: A comparative study of face-to-face communication and virtual communication by daily life behavior
abstract: The influence of Covid-19 has increased the opportunity to communicate online. In addition, there are many technologies and services that aim to virtualize our daily lives. On the other hand, the extent to which they cover our daily lives has not been discussed much. In this research, we aim to clarify which areas are untapped in the virtualization of people's daily activities and why they cannot be explored, and then to achieve virtualization in these untapped areas. Prior to the development of the technology, we conducted a survey of 19 different types of daily life behaviors to determine the differences between face-to-face and virtual communication. As a result, lack of communication and sense of presence were frequently cited as disadvantages of virtualization. In addition, among the elements that make up the sense of presence, differences in real time and synchronization were found to be small, while large differences were found in the sense of self-presence and interactivity. In the future, based on the findings from this survey, we will consider various approaches to achieve virtualization in untapped areas.
language of the presentation: Japanese
発表題目: 生活行動ごとの対面・バーチャルによるコミュニケーションの比較検討
発表概要: Covid-19の影響でオンラインでコミュニケーションを取る機会が増えた。また、それに伴い日常生活のバーチャル化を目的とした多くの技術やサービスが存在する。一方でそれらが日常生活をどれだけカバーしているかについてはあまり議論されてこなかった。そこで本研究では人の日常生活行動のバーチャル化においてどの領域が未開拓か,なぜ開拓できないかを明示したうえで、未開拓な領域でのバーチャル化を達成することを目的としている。そのため技術開発に先立ち、人が互いにコミュニケーションを取る際に対面とバーチャルでどのような違いがあるのか19種類の生活行動に対して調査を行った。その結果、バーチャル化による欠点では,コミュニケーション不足や臨場感の不足が多く挙げられた.また、臨場感を構成する要素のうち実時間性や同期感についてはその差が小さく見られる一方で自己存在感やインタラクティブ感などは大きな差異が考えられた。今後はこの調査から得られた知見を踏まえ、未開拓な領域でのバーチャル化を達成するために様々なアプローチを検討する。
 
松本 敢大 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, 金谷 重彦, 諏訪 博彦
title: Health status prediction using daily living activities data and heart rate information for the elderly
abstract: In order to extend the healthy life span of the elderly, there is a need for a system that can measure QoL (Quality Of Life) sequentially and provide feedback tailored to individuals and situations. However, conventional QoL measurement methods, such as questionnaires and interviews, are difficult to conduct frequently, and feedback is limited to encouraging behavioral changes in general lifestyle behaviors such as exercise and dietary habits. Therefore, the purpose of this study is to construct a system that estimates QoL from biometric data for each daily living activities, identifies behaviors that decrease QoL, and then encourages behavior change. In order to examine the feasibility of estimating QoL from biometric data for each of the daily living activitiess, we analyzed a dataset of about 30 days consisting of data on the daily living activitiess and heart rate of nine subjects from five families and a subjective questionnaire with three questions in the morning and four questions at night for measuring QoL. As a result of calculating the Lorenz plot area from the heart rate data for each behavior and comparing the relationship with the QoL questionnaire, it was confirmed that there was a relationship between the Lorenz plot area and the questionnaire results for several behaviors. As a result, it was confirmed that there was a relationship between the Lorenz plot area and the questionnaire results in several behaviors. In addition, a model to estimate each QoL questionnaire was constructed by random forest using the daily living activities record and the Lorenz plot area as the characteristic quantity. This suggests the possibility of estimating QoL from biological data for each daily life activity. In the future, we will improve the accuracy of the estimation and study the behavior change method.
language of the presentation: Japanese
発表題目: 高齢者対象にした家庭内での行動データと心拍情報を使った健康状態予測
発表概要:高齢者の健康寿命延伸のために,逐次的にQoL(Quality Of Life)を計測し,個人や場面に合わせたフィードバックを行うシステムが求められている. しかし,従来のQoL計測手法であるアンケートや問診は高頻度で行うことが困難であり,フィードバックも運動や食生活など生活行動全般に対する行動変容を促すものとなっている. そこで本研究では,日常生活行動ごとに生体データからQoLの推定を行い,QoLを低下させる行動を特定したうえでその行動変容を促すシステムを構築することを目的とする. 日常生活行動ごとに生体データからQoLを推定することの実現性を検討するために,5家庭・9人の被験者の生活行動および心拍のデータと,QoLを計測するための朝3問,夜4問の主観的なアンケートからなる約30日間のデータセットを分析した. 生活行動ごとに心拍データからローレンツプロット面積を算出しQoLアンケートとの関係を比較した結果,複数の行動でローレンツプロット面積とアンケート結果に関係があることが確認された. また,生活行動記録とローレンツプロット面積を特徴量とし,各QoLアンケートを推定するモデルをランダムフォレストにより構築した結果,各質問に対し約60%以上の正答率を得ることができた. これにより日常生活行動ごとに生体データからQoLを推定することの可能性が示唆された.今後は,推定精度の向上と行動変容手法の検討を行う.
 
工藤 創大 M, 1回目発表 計算システムズ生物学 金谷 重彦, 安本 慶一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘

title: Fine Tuning Based Approach for Sleep Stage Classification with Acceleration and Heart Rate Data  from a Consumer  Wearable Device


abstract: The current gold standard for clinical sleep monitoring is polysomnography. It is performed in clinics or hospitals and requires complex equipment setup and manual scoring by specialists, making it inappropriate for routine monitoring. For long-term sleep monitoring, actigraphy, which can detect motion is used, however its accuracy is known to be limited. Currently, consumer wearable devices are getting popular and are attracting attention as sleep monitoring devices, but their use for clinical and research purposes is limited because their algorithms are not disclosed. To increase the transparency, a machine learning model for 3-class (wake, REM, Non-REM) sleep stage classification using consumer wearable devices has been proposed so far. In this study, for the purpose of practical 4-class (wake, REM, N1/N2, N3) classification, we proposed a GRU-based machine learning model that considers temporal information during whole sleep, and pre-trained it using a more abundant dataset of clinical and research sensors. The MESA dataset (1743 subjects) for pre-training, including actigraphy data, photoplethysmography-derived heart rate data and ground truth sleep stage scored by experts, and the open-source Apple Watch dataset (31 subjects) having raw acceleration data, heart rate data and sleep stage score were used.


 
中川 翔太 M, 1回目発表 計算システムズ生物学 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘
title:Create a Generalized Model of Health Information from Chest X-Ray Images by Embedding Using Deep Learning.
abstract: Chest X-ray images are one of the most common diagnostic techniques used in routine clinical practice around the world because they can diagnose various diseases (pneumonia, lung cancer, pleural thickening, spinal scoliosis, etc.). In order to support diagnosis, research has been conducted to develop medical diagnosis support systems, and in recent years, research to introduce deep learning into these systems has been gaining attention. However, the deep learning models proposed so far are not capable of comprehensively evaluating multiple diseases. In this study, we propose to create a generalized model of health information by embedding chest X-ray images using an unsupervised learning model to address this problem.
language of the presentation:Japanese
発表題目:深層学習を用いた胸部X線画像の埋め込みによる健康情報の汎化モデルの作成
発表概要:胸部X線画像は様々な疾患(肺炎、肺がん、胸膜肥厚、脊椎側湾症など)を診断することができるため、世界中の日常的な臨床現場で使用される最も一般的な診断技術の1つである。診断をサポートする為に、医療診断支援システムの開発研究が行われており、近年、これに深層学習を導入する研究が注目を集めている。しかし、これまでに提案された深層学習モデルでは複数の疾患を包括的に評価することができない。本研究ではこのような問題に対して教師なし学習モデルを使用した胸部X線画像の埋め込みで健康情報の汎化モデルの作成を提案する。
 

会場: L3

司会: 小林 泰介
安川 浩貴 M, 1回目発表 知能コミュニケーション 中村 哲, 渡辺 太郎, 品川 政太朗
title: Extracting persona information by using contrastive learning and applying it to the neural dialogue model
abstract: In recent years, dialogue models using neural networks have been able to generate natural sentences at a level comparable to that of humans. On the other hand, some previous studies reported that highly-scored sentences in human evaluation reflect speaker characteristics. This suggests that it is not only the naturalness of the sentences that is important in dialogue. Despite the importance of speaker characteristics, the model performance is still limited to reflect it in response generation. In this paper, we propose a method to generate responses that reflect the speaker's characteristics by constructing a model that learns vectors representing the speaker from sentences describing the speaker's characteristics using contrastive learning and inputting the acquired vectors to a dialogue model. And in this presentation, we focus on the part of the proposed method that extracts the vectors that represent speaker information.
language of the presentation: Japanese
発表題目: Contrastive learningを用いたペルソナ情報の抽出とニューラル対話モデルへの適用
発表概要: 近年のニューラルネットワークを用いた対話モデルは,人間と同等のレベルで自然な文章を生成できるようになってきた.その一方で,先行研究では人手評価が高い返答は,話者の性質を反映している文章であることが報告されている.このことから対話において重要となるのは文章としての自然さだけではないと言える.このように、話者の情報を反映する事は重要であるにも関わらず,返答文生成での話者情報の反映という点では、モデルの性能が未だ限られている。そこで本研究では,話者の性質を記述した文章から話者を表現するベクトルをContrastive learningによって学習するモデルを構築し,獲得したベクトルを対話モデルへと入力する事で話者の性質を反映した返答を生成する手法を提案する.本発表では,提案した手法の内の話者情報を表現するベクトルを抽出する部分を中心に紹介する.
 
安藤 晶 M, 1回目発表 ソーシャル・コンピューティング 荒牧 英治, 中村 哲, 若宮 翔子, 矢田 竣太郎, 大竹 義人
title: A Study on the Integrity of Nursing Practice Using Nursing Records
abstract: Nursing records are medical documents to prove nursing practice, ensure continuity and consistency, and improve evaluation and quality. Therefore, nurses produce a huge amount of records every day. Few studies have been reported on nursing records in the Natural Language Processing field, because these documents contain a lot of patient's personal information and require medical knowledge to prepare the training data. On the other hand, in clinical environments, auditing of medical documents such as nursing records and many other document verification tasks are done manually. In this study, we aim to develop a nursing NER (Named Entity Recognition) system to automate the auditing work by the inetgrative confirmation of nursing practice.
language of the presentation: Japanese
発表題目: 看護記録を用いた看護実践の整合性調査
発表概要: 看護記録は看護実践の証明,継続性および一貫性の担保,評価及び質の向上を目的とした医療文書であり,看護師により日々膨大な記録が作成されている.看護記録は患者の個人情報を多く含み,更に教師データの作成に医学的知識を要するため言語処理分野での研究例がまだ少ない.一方で,医療の現場では看護記録を含む医療文書の監査をはじめとする多くの文書確認作業が人手で行われている.そこで本研究では看護実践の整合性調査による監査業務の自動化を目的として,看護表現抽出器の開発を行う.
 
杉原 健斗 M, 1回目発表 大規模システム管理 笠原 正治, 藤川 和利, 笹部 昌弘, 原 崇徳
title: A Survey on Virtual Network Function Placement Schemes to Improve Availability and Reliability of Service Paths
abstract: Network functions virtualization (NFV) can deploy diverse and flexible services by executing network functions at generic hardware as virtual network functions (VNFs) instead of dedicated hardware. A certain network service can be represented as a sequence of VNFs called a service chain. Given the locations of functions, service chaining tries to find a service path from an origin and a destination while executing the VNFs at intermediate nodes in required order. The locations of functions will not only affect the optimality of service path but also its availability and reliability under failures of nodes, links and VNFs. In this presentation, we introduce existing studies on VNF placement to improve the availability and reliability of service paths.
language of the presentation: Japanese
 
平出 託海 M, 1回目発表 大規模システム管理 笠原 正治, 井上 美智子, 笹部 昌弘
title: A Mathematical Model of Blockchains Considering Dependencies of Fees, Approval Delay, and Security

abstract: In a Bitcoin-type blockchain, the minor nodes prioritize the transactions with higher fees to be included in the block earlier. Therefore, end users are interested in the fee required to process a transaction with an acceptable delay. Miners are interested in increasing the number of transactions in a block in order to increase the revenue from fees. An increase in block size tends to increase block approval processing and network propagation delays, which not only reduces end-user utility, but also may lead to a vulnerable security situation due to frequent forking. In this study, we consider a mathematical model that expresses the dependency among the three factors of transaction fees, approval processing delay, and security, and verify its validity. *
 
北澤 太基 M, 1回目発表 情報セキュリティ工学 林 優一, 岡田 実, 中島 康彦, 藤本 大介, Youngwoo Kim
title: A Study on the Effect of the Via Placement in the Connector Mounting Area on Electromagnetic Emission
The unintentional electromagnetic (EM) emission from the connectors becomes a problem because it affects the other devices and wireless communication. The main factor of EM emission is that the noise generated in the printed circuit boards (PCB) propagates to the connectors. Therefore, there is a demand to investigate the mechanism of these noise emissions and countermeasures against them. Conventional studies have mainly focused on the noise caused by integrated circuits and transmission lines on PCBs. However, there is not enough discussion about the connector mounting area. In the connector parts, because it is difficult to design ideally due to density and complexity, the optimal design is required, the placement of ground vias for noise suppression could cause noises. In this study, we investigate the relationship between the placement of ground vias in the connector mounting area and noises which cause EM emission.
Japanese
コネクタ相互接続部付近のビア配置が電磁放射に与える影響評価
USBやHDMIなどのコネクタは,設計者が意図しない電磁放射を発生させ他機器や無線通信に影響を与えるため問題となっている.電磁放射の主な要因は,基板で発生するノイズがコネクタまで伝搬するためである.よって,ノイズの発生原因や対策などを明らかにする必要がある.従来研究では,伝送路やICなどで発生するノイズの発生メカニズムについて多く検討されていたが,基板とコネクタの相互接続部分における検討は十分にされていない.相互接続付近は実装密度が高く理想的な設計を行うことが難しく,最適な設計手法を示す必要がある.中でもノイズ対策のために実装するGroundビアの配置位置とノイズの関係性は明らかになっていない.本研究では,コネクタと基板の相互接続部に存在する Ground ビアの配置位置と放射の要因となるノイズの関係性について検討した.
 
PHAN THUY PHUONG M, 1回目発表 ネットワークシステム学 岡田 実, 林 優一, 東野 武史, DUONG QUANG THANG, Chen Na
title: Electric magnetic fields reduction for inductive power transfer system using parallel line feeder
abstract: RF driven two parallel wires have been proposed as a simple transmitter coil for delivering wireless power to receivers moving along a line. Unfortunately, this system has the issue of leaking a strong magnetic field to surrounding space. This research therefore examines the possibility of using three parallel wire structures to reduce the spread magnetic field. In the three-wire system, the center wire carrying a current has a reverse direction with the currents in two remaining wires. The magnetic field generated by all the three wires can be canceled by each other in the outer region. As the theoretical result, the new system can bring the expected effect of a significant reduction in leaked magnetic field strength compared to the conventional two-wire system. Currently, theoretical analysis for the three-wire system has been finished. Extensive EM simulations and experiments will be carried out in the next steps to confirm effectiveness of this new design.
language of the presentation: English