| 中村 優太 | M, 1回目発表 | 大規模システム管理 | 笠原 正治 |
|
title: Capability-based Access Control using Smart Contract for IoT
abstract : It is reported that, by 2020, over 30 billions of IoT devices, like household appliances, automobiles, medical and industrial equipment, will be connected into the Internet. Although these devices make our life smarter and more convenient, once accessed illegally, they may incur significant safety issues like personal information leakage and traffic accidents. Access control is effective to prevent the illegal access to these devices. Access control for IoT systems needs high scalability to handle massive access requests concurrently, and also high flexibility to meet the diverse requirements of these requests. In this research, we propose an capability-based access control framework based on the blockchain technology to meet the above requirements. language of the presentation: Japanese | |||
| 西 洋平 | M, 1回目発表 | 大規模システム管理 | 笠原 正治 |
|
title: On Impact of Alleviation of Tit-for-Tat Constraint between Peers Belonging to Identical Organizations on P2P File Distribution
abstract: Peer-to-Peer (P2P) file distribution can solve the drawbacks of conventional client-server file distribution, e.g., load concentration and single point of failure, by allowing clients called peers to upload retrieved fragments of the whole content, i.e., pieces, to other peers. However, the success of P2P file distribution highly relies on the cooperative behavior of peers. Due to communication overhead, some peers will not be willing to upload pieces to others, which are called free riders. Tit-for-Tat (TFT) strategy in game theory can alleviate such free riding behavior by encouraging equivalent exchange of pieces among each pair of peers. In recent years, the optimality of P2P file distribution under strict TFT constraint has been analyzed. In this paper, we focus on the fact that the communication overhead inside an organization is much less than that between different organizations, and thus remove the TFT constraint among peers belonging to identical organizations. We find that The minimization of average file download time over such P2P file distribution with alleviated TFT constraint can also be modeled as Integer Linear Programming (ILP), as in the existing work. Through numerical evaluations, we show that alleviated TFT constraint can shorten average file retrieval time of peers with slight performance degradation. language of the presentation: Japanese | |||
| 西田 雄亮 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基 |
|
title: Detecting Split-Personality Malware Fingerprinting Evidence of Configuration
abstract: Due to spread and the obfuscation of Malware, analysts use automated tools which observe behavior of suspicious file. But many malwares detect analysis environments and behave innocuously. Thus, automated tools observe wrong behavior and analysts take extra time to analyze it. In this research, we aim to propose a method that can detect it and mitigate its behavior. language of the presentation: Japanese | |||
| 西本 宏樹 | M, 1回目発表 | コンピューティング・アーキテクチャ | 中島 康彦 |
|
title: Design and development of VBGMM accelerator
abstract: In recent years, machine learning techniques have been developed with the improvement of CPU and GPGPU performance, and it has been used for analyzing big data, etc. However, for CPU and GPGPU, processing can not catch up anymore, In addition, one of the clustering methods by machine learning is a method called variational mixture Gaussian distribution that uses variational reasoning that uses the concept of Bayesian inference There is a characteristic that it is difficult to over learn this method compared with mixed Gaussian distribution which does not use variational reasoning.We decided to design and develop accelerator of this variational mixed Gaussian distribution. language of the presentation: Japanese 発表題目: 変分混合ガウスモデルアクセラレータの設計・開発 発表概要:近年, CPU や GPGPU 性能の向上とともに機械学習技術が発達し，ビッグデータの解析等に用いられる ようになった.しかし，増加する要求計算量に対し，CPU や GPGPU ではもはや処理が追いつかず，タスクごとに専 用のアクセラレータが設計されることも多くなった.また，機械学習によるクラスタリング手法の一つに，ベイズ推 論の概念を利用する変分推論を用いた変分混合ガウス分布という手法がある.この手法は変分推論を利用しない混合 ガウス分布に比べて過学習を行い難いという特徴がある.我々はこの変分混合ガウス分布のアクセラレータの設計・開発を行うことにした. | |||
| 萩森 大貴 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清 |
|
title: Influence of active kinesthetic illusion by tendon vibration stimulation on somatosensory sensation during elbow joint movement
abstract: We develop that Virtual Reality (VR) interfaces without actual Movement. Input of the VR interfaces is the same sense as reality. For example, the VR legs begin to move, if you think that you want to move the real legs. The VR interfaces reduce a space and realize advanced VR experience. In this study we investigated the change of elbow angle with vibration is smaller than the angle without it by motion illusion of applying tendon vibration when bending elbow joint angle perceived by somatosensory sensation. language of the presentation: Japanese | |||
| 原田 康嗣 | M, 1回目発表 | 情報基盤システム学 | 藤川 和利 |
|
title:[paper introduction] When UAVs Ride A Bus: Towards Energy-efficient City-scale Video Surveillance
abstract: Unmanned aerial vehicles (UAVs) are being increasingly deployed for next-generation surveillance of urban areas, driven by availability of off-the-shelf devices and reducing costs. However, most readily deployable solutions today are based on a single UAV-ground station link, with limited flying time imposed by the on-board battery. This presentation introduce their proposal method that uses public transportation for re-charging and traveling to the expected points of interest. Finally , I will show the direction of my research plan. language of the presentation: Japanese | |||
| 平野 陽大 | 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 |
|
title:Multi-modal route search with multi objective GA for decision making support on user's resource consumption
abstract:Recently, the number of people who go on a trip are increasing. However,it is difficult to decide plans of trip, because there are many factors to consider in the tour. For example, we are concerned about our context such as money, stamina, and time. Thus, we suggest a recommender system considering and optimizing consumption of user resources.The problem of planning a trip taking into account users resource is a multi-objective optimization problem with multiple objective functions. Thus, it is difficult to derive the optimal solution in practical time. Therefore, we use multi objective GA and get solutions as pareto semi-optimal solution. And also, we consider the best way to indicate pareto optimal solutions for decision support. language of the presentation:Japanese | |||
| 弘世 和久 | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦 |
|
title: Molecular pathway identification for Schizophrenia and Bipolar Disorder based on marker metabolites
abstract: Schizophrenia and bipolar disorder are common and complex psychiatric disorders. Although there are a lot of hypotheses that these disorders occur, molecular mechanism is unknown. In order to understand protein-protein interaction of schizophrenia and bipolar disorder, we construct protein-protein interaction network from marker metabolites in disorders and analyze with network clustering, DPclusO. language of the presentation: Japanese | |||
| 岡本 真由子 | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title: Generative adversarial network speech synthesis for speech entrainment in Human-Machine Interaction
abstract: In human-human communication, entrainment phenomenon often occurs: two speakers tend to co-adjust their speaking style to ease communication. This may include convergence in vocabulary, speaking habit, speaking rate, loudness, pitch, and so on. Existing study found that when dyads' speech rates converged more strongly from the beginning to the end of the conversation, they were more likely to cooperate with each other. It is therefore desirable that synchronization in the human-machine interaction like human-human communication also occurs. In this research, we aim at the speech rate convergence between the machine and the partner. In this first stage, we focus on developing a speech synthesis system that can change the speaking rate according to the partner. Specifically, we construct generative adversarial network speech synthesis (GAN-TTS) that can produce speech with a various speaking rate. language of the presentation: Japanese | |||
| 堀口 博 | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Entity linking and acronym expansion in clinical text
abstract: In the ICD code automatic classification task of MIMIC 3, embedding with the context taken into account was not used in previous research, and information possessed by the label itself was not taken into consideration. In this presentation, we will announce how we deal with abbreviations in diagnostic notes and do text classification, while using embedding of label and label information for text classification. Mainly, partial task of text classification(abbreviation expansion and entity linking) will be described. language of the presentation: Japanese | |||
| 本多 克 | M, 1回目発表 | 環境知能学 | 萩田 紀博 |
|
title: Atmosphere sharing method using TV chat agent
abstract: In this research, we propose and evaluate the atmosphere sharing method using TV chat agent for the purpose of improving the motivation of users' continuous dialogue. This method proposed an agent expression method for sharing the atmosphere when watching TV based on the excitement section determination method and the excitement level determination method using the SNS comment in the previous research. By expressing according to the excitement section and the excitement level, it is possible to give the user an impression such as unity and sympathy, and it is expected to improve motivation for continuous dialogue. At this stage, it is the stage where the implementation of the atmosphere sharing method is completed and the evaluation is carried out. language of the presentation: Japanese | |||
| 三﨑 慎也 | 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:Design and Evaluation of Intermittent Operation Module with Energy Harvesting for In-Network Processing Sensor Nodes
abstract:The large indoor space such as shopping malls where many people gather requires a service which monitors temperature and person density, detect anomaly in temperature and keep appropriate air conditioning over the space. In order to realize such a service, it is necessary to install many environmental sensors, human detection/tracking sensors, etc. However, sensors need a battery or power supply that are costly for maintenance and/or deployment. On the other hand, analysis of data output by sensors is often concentrated in clouds, servers, etc., but such a cloud-based system wastes communication resources. Our goal is to solve the above problems by operating sensor nodes by energy harvesting and performing data processing in the mesh network formed by those sensor nodes. In our previous study, we developed a module that performs an intermittent operation by charging and discharging a capacitor by using a solar panel. However, it is designed to start after the sensor node is charged with a sufficient amount of power required to complete a series of sensing and data transmission operations. Therefore, the interval of the intermittent operation changes due to environmental changes (changes in the charged amount due to brightness in the environment), making it difficult to operate at constant intervals. In this study, we design and evaluate a new energy harvesting module consisting of devices capable of intermittently operating the microcomputer at an interval specified by user and the battery charging device with electric power generated from sunlight. language of the presentation: Japanese | |||
| 松岡 潤樹 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 |
|
title: Improving Sample Efficiency of Reinforcement Learning Using Learned Policies
abstract: Learning of the robot movement with the reinforcement learning (RL) has attracted attention, and improvements of various RL methods have been carried out intensively. The learning with the RL methods generally costs too long learning time, which makes the re-learning when performing another task extremely inefficient. Therefore, a RL method that can re-learn more efficiently has been required. Then, we focused on the compositionality of policies of Soft Q-Learning (SoftQL) proposed by Haarnoja et al. With SoftQL, it is possible to compose multiple already-learned policies and to execute compound tasks efficiently. However, there are the problems that sampling of actions and learning algorithm are difficult due to the continuity of action space. In this presentation, I will introduce SoftQL and describe the idea for developing more practical RL method with the compositionality of policies. language of the presentation: Japanese | |||
| 松本 彩香 | 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: Consideration on Motivation Maintenance for Sustainable Training
abstract: The number of elderly people who are under rehabilitation after stroke or physical injury is increasing. Continuing rehabilitation directly benefits them by improving quality of life (QOL) and reducing burden of people around them, but many elderly give up before they recover because rehabilitation is hard and long-term. Motivation is important factor in the rehabilitation. There are already a lot of rehabilitation support system, and gamification method is often used for the system. Gamification leverage people's natural desires for learning, mastery, competition, achievement, so that the task with gamification fascinate people even if it is boring without gamification. However, gamification is used for rehabilitation exercise only, although rehabilitation takes place not only in exercise, but also in activities of daily living and in social participation.We aim to develop digital support system which motivate elderly in rehabilitation three stage; exercise, activities of daily living, and social participation, to continue rehabilitation and verify if our system work well through user study with elderly people who are under rehabilitation. language of the presentation:Japanese | |||
| 三上 徹朗 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 |
|
title: Constructing device of hyperspectral imaging in near infrared for analyzing pre-modern Japanese works.
abstract: Visualization using near infrared images for ownership stamps and ink characters relating to Pre-modern Japanese works is important.Near infrared images of 800~1100 nm have also been used for analysis of works. If we can analyze using near-infrared image of 800~2500 nm with long wavelength and high wavelength resolution, visualization of stamps and ink characters can be expected from the distribution of chemical substances. Separation due to differences in chemical composition of stamps and ink characters can be expected. We constructed a hyperspectral imaging device by combining a spectrometer capable of acquiring spectral information in the entire near infrared range and automatic stages. language of the presentation: Japanese 発表題目: 古典籍の解析のための近赤外領域における高解像度分光画像計測装置の構築 発表概要: 古典籍に関する蔵書印や墨文字に対して近赤外画像を用いた可視化は重要である。これまでの古典籍の解析にも800~1100nmの近赤外線画像が用いられてきた。更に長波長で高波長分解能な800〜2500nmの近赤外画像を使って解析できれば、蔵書印や墨文字の化学的な組成の違いによる分離が期待できる。我々は、近赤外線全域の分光情報を取得できる分光器と自動ステージを組み合わせて、計測装置を構築した。 | |||
| 牧 泰宏 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 |
|
title: Compensation of time-variant delay for networked control system
abstract: In the networked control system, the devices mutually transmit information using shared resources such as internet lines. Therefore, a reduction in transmission rate and band limitation occur, and information transmission irregularly delays. For such a system, we propose a gain switching observer based on time stamp information in order to compensate the delay and perform appropriate control. language of the presentation: Japanese | |||
| 三野 智貴 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子 |
|
title: Physical Unclonable Function for Device Authentication and Secret Key Geneation
abstract: In recent years, electronic devices of fake and copy are considered problematic. If they flow in circulation, the brand image of enterprises will be damaged. Also, the electronic systems which are used fake devices will decrease reliability. It has a device authentication to protect from the problems. Mainstream of the way is to place a ID on ICs of Non-volatile memory. However, it takes high cost to use the way with high security. Therefore, it required device authentication way with high security and low-cost. As one of the way to realize it, there is Physically Unclonable Function(PUF) which exploits process variations. In this presentation, I will introduce a paper that is related PUF. First, I will explain about Ring Oscillator PUF that made a proposal in the paper. Then, I will explain about my research. language of the presentation: Japanese | |||
| 宮本 知弥 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 |
|
title:Trace Paths Search for Tacticle Object Shape Exploration
abstract: To estimate the shape of the object, the point search is an inefficient touch and retract. As an improvement method, it is conceivable to adapt trace paths search such that a human touches an object to the robot. This realizes more efficient paths search. In the conventional method, it is possible to estimate the shape of a convex object with trace paths search, but it cannot be a concave object. So it is not a general paths search. Therefore, in this research, we propose a method of trace paths search to estimate the shape of a more general object including a concave shape.br /> language of the presentation: Japanese | |||