| 宇恵 勝紀 | M, 1回目発表 | 数理情報学 | 池田 和司☆ | 作村 諭一 | 川鍋 一晃(客員） | 森本 淳(客員） | 福嶋 誠 |
|
title: *** The influence of brain dynamics by neuromodulator ***
abstract: *** Although the mechanism of the generation about brain activity is described in several points, it is little done to try to explain brain activity with mathematical models on neuromodulators. In this presentation, I will introduce research that explained the brain dynamics from the structure of the brain as one of the mechanism of the generation about brain activity. Then, I will introduce the research that reproduced the brain dynamics using distribution data of serotonin receptors as one of the previous research. And I will describe my plan about how I expand the previous research. *** language of the presentation: *** Japanese *** 発表題目: *** 神経修飾物質による脳ダイナミクスの影響 *** 発表概要: *** 脳活動の発生メカニズムは様々な観点から数理モデルを用いて提唱されているが、神経修飾物質を数理モデルに組み込んで脳活動を説明する試みはほぼ成されていない。本発表では、脳活動発生メカニズムの1つとして、脳構造からダイナミクスを説明した例を挙げ、先行研究として脳でのセロトニン受容体(5-HT2A receptor)の分布データを用いた脳ダイナミクスを再現した研究を紹介する。そして、先行研究をどのように発展させていくかの計画を述べる。 *** | |||||||
| 岩田 晟 | M, 1回目発表 | 自然言語処理学 | 渡辺 太郎 | 中村 哲 | 進藤 裕之 | ||
|
title: Japanese Zero-Pronoun Detection with BERT
abstract:Zero-pronouns, where subject and object nouns are not explictly expressed, are often found in Japanese. In Natural Language Processing (NLP), this noun deletion negatively affects the performance of application tasks such as machine translation. In this research, we propose a detecting method for zero-pronouns using a pre-training model BERT, widely used in various NLP tasks. language of the presentation: Japanese 発表題目:事前学習モデルBERTを用いたゼロ代名詞検出 発表概要:日本語や中国語では文中の主語や目的語などの名詞がしばしば省略されることがある．そのように省略された名詞をゼロ代名詞という．そのような名詞の欠落は機械翻訳などの応用タスクの性能に影響を及ぼす．本研究では，文章中のゼロ代名詞を近年，自然言語処理分野で活躍している事前学習モデルBERTを用いた手法を検討する． | |||||||
| 島田 尚道 | M, 1回目発表 | ソフトウェア工学 | 松本 健一 | 安本 慶一 | 石尾 隆 | 畑 秀明 | Kula Raula Gaikovina |
|
title:
Analysis of the impact of GitHub Sponsors donations to individual developers on projects
abstract: GitHub Sponsors were implemented last year with the aim of supporting OSS developers.GitHub Sponsors is a system to provide continuous financial support as a donation to developers who contribute to OSS.Prior to the implementation of GitHub Sponsors, the mainstream way to donate to a project was through external platforms. In contrast, GitHub Sponsors is a donation format that allows you to focus more on individual developers than the traditional method.Past research has shown that projects that receive donations become more active, and that issues issued by donors to such projects are closed preferentially.However, past research has focused on donations to projects, and it is not yet clear what impact individual-focused forms of donation, such as GitHub Sponsors, have on projects.Therefore, we investigate the impact of individual donations by GitHub Sponsors on the project. language of the presentation:Japanese 発表題目: GitHub Sponsorsによる個人への寄付がプロジェクトに与える影響の分析 発表概要: OSSの開発者を支援することを目的として，昨年からGitHub Sponsorsが実装された．GitHub SponsorsとはOSSへ貢献する開発者への寄付として継続的な金銭支援ができる仕組みである．GitHub Sponsors実装前は外部プラットフォームを通じて，プロジェクトへの寄付を行うことが主流であった．対してGitHub Sponsorsは従来の方法に比べて，より開発者個人に焦点をあてて寄付を行うことができる寄付形態をとっている．過去の研究では，寄付を受けたプロジェクトは活発になることや，当該プロジェクトへの寄付者が出したissueは優先的にクローズされることが分かっている．しかし過去の研究はプロジェクトへの寄付に関する研究であり，GitHub Sponsorsのような個人に焦点をあてた寄付形態がプロジェクトにどのような影響を与えるかはまだ定かになっていない．そこでGitHub Sponsorsによる個人への寄付がプロジェクトに与える影響を調査する． | |||||||
| OSMANI SHAIRA | M, 1回目発表 | 大規模システム管理 | 笠原 正治 | 松本 健一 | 笹部 昌弘 | 張 元玉 | |
| title: Evaluation of Block Withholding Attack and Its Detection Possibility abstract: In the Bitcoin system, transactions are recorded in a distributed ledger called blockchain, which is a sequence of blocks. Some special nodes called miners try to create a new valid block in order to acquire rewards (i.e., new Bitcoins) by solving cryptographic puzzles with certain difficulty (i.e., network difficulty). This process is called proof of work (PoW) and it requires a huge number of hash calculations. To acquire the rewards while suppressing the electricity and investment costs, multiple miners tend to form a group called a pool to collaboratively conduct PoW. The manager of the pool, i.e., pool manager, divides the original PoW task into multiple sub-tasks and allocates them to the member miners. It also sets the local difficulty of PoW (i.e., pool difficulty), which is easier than the network difficulty, to confirm the contribution of members. Each member is requested to report their finding blocks and shares, which only satisfy the pool difficulty condition, to the pool manager. The pool manager distributes rewards to members according to their contribution. It has been pointed out that some malicious miners can sabotage the mining process and gain more rewards by hiding found blocks. This attack is called block withholding attack. In this presentation, we focus on the two kinds of ratios: 1) The ratio of pool difficulty to network difficulty and, 2) the ratio of the number of reported blocks to that of reported shares for each member. These ratios can be calculated by the pool manager only using the observable information. For honest miners, these two ratios would converge to the same value. On the other hand, for malicious miners hiding some blocks, the second ratio tends to be lower than first ratio. As a result, we expect that the difference between these two ratios can be an indicator to detect malicious miners. Since the convergence would require some time, we examine the possibility of detection under different control intervals through simulation experiments using the modified version of the existing simulator PoolSim. language of the presentation: English | |||||||
| 冨田 周作 | 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 | 笠原 正治 | 諏訪 博彦 | 中村優吾 | |
|
title: A Study on Sharing Method of Tourism Object Recognition Model by Federated Learning Based on DTN
abstract: In recent years, AI and IoT-based tourism support services are becoming more and more popular, and one of them is to support tourists by using object recognition to recognize a variety of contexts at various tourist sites. Although Federated Learning is considered to be a suitable method to train an object detection model with privacy data, it may increase the communication cost if the edge devices required for training are widely distributed. In this study, we propose a method for sharing an object detection model between tourist sites by using a delay tolerant network (DTN) to reduce the communication cost of federated learning with edge devices only. language of the presentation: Japanese 発表題目: DTNに基づいたFederated Learningによる観光オブジェクト認識モデルの共有手法の検討 発表概要: 近年、AIやIoTを用いた観光支援サービスが普及しつつあり、その1つに物体認識による様々な観光地での多様なコンテキスト認識を活用した観光客への支援が考えられる。プライバシを含む観光地のデータを観光地間で物体認識モデルに学習させるにはFederated Learningが適すると考えられるが、学習に必要なエッジデバイスが広範囲に多く分布する場合、通信コストの増加が懸念される。本研究では、DTN(遅延耐性ネットワーク)によるエッジデバイス間のみの通信でFederated Learningを実施することで通信に関わるコストを削減し、観光地間で物体認識モデルを共有する手法を検討する。 | |||||||
| 橋本 律紀 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一 | 藤川 和利 | 安本 慶一 | 藤本 大介 | Youngwoo Kim |
|
title: A Study on the Mechanism of Randomness Degradation in RO-based TRNGs by the Frequency Injection Attacks
abstract: True Random Number Generators (TRNGs) based on ring oscillators (ROs) are widely employed due to their simple structure. Frequency injection attacks have been pointed out as a security threat against RO-based TRNGs to degrade their randomness by intentionally injecting electromagnetic waves of a specific frequency. The frequency varies depending on the oscillation frequencies of ROs implemented in the victim TRNG. However, the mechanism of the randomness degradation of a RO-based TRNG by frequency injection has not elucidated. Thus, there is no effortless method to choose the injection frequency for an arbitrary RO-based TRNG. Therefore, this study examines the mechanism of randomness degradation of a RO-based TRNG caused by frequency injection by frequency and time domain measurements. In frequency domain measurement, influence of frequency injection on the oscillation frequency of TRNG are observed by frequency sweep. In time domain measurement, jitters of oscillation frequencies of ROs, which are the source of the randomness of RO-based TRNGs, are observed to see if the injected frequency degrades its randomness. The study also covers possible countermeasures for the frequency injection attacks. language of the presentation: Japanese | |||||||
| 長瀬 康斗 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 | 加藤 博一 | 舩冨 卓哉 | 田中 賢一郎 | |
|
title: Passive Ranging Using Multichannel Far Infrared Measurements
abstract: In the field of computer vision, distance measurement technology is used in various applications such as autonomous driving and robot control. Currently, stereo camera and time-of-flight technologies are the mainstream, but there are limitations such as the need for textures of the target and the known light sources. In this study, we propose a method for estimating distance by textureless and passive measurement using far-infrared measurement. The proposed method uses Lambert-beer's law to model the atmospheric attenuation of far-infrared rays emitted from a target, and shows that the distance, temperature, and emissivity of the target can be estimated at the same time by performing multi-channel measurement. In order to realize the proposed method, we established an experimental system and determined the attenuation coefficient of the atmosphere. By using the measurement results and the proposed model, it was suggested that the distance to a target can actually be estimated. In the future, we aim to improve the accuracy so that it can apply many scenes. language of the presentation: Japanese 発表題目: マルチチャンネル遠赤外線計測を利用した受動的な距離測定 発表概要: コンピュータービジョンの分野における距離測定技術は，自動運転やロボットの制御など様々なアプリケーションに用いられている． 現在ではステレオカメラやtime-of-flightなどの技術が主流であるが，測定対象にテクスチャが必要であったり，既知の光源を必要とするなどの制限がある． 本研究では遠赤外線計測を利用したテクスチャーレスかつ受動的な計測で距離推定を行う手法を提案する．提案手法ではLambert-beerの法則を使って 対象から放出される遠赤外光線の大気減衰をモデル化し，マルチチャンネルな計測を行うことで，対象の距離，温度，放射率が同時に推定できることを示す． 提案手法を実現するために，実験系を確立し，大気の減衰係数などの計測を行った．計測結果と提案したモデルを用いることで，測定対象との距離が実際に推定可能あることを示唆した． 今後は多くのシーンに対応できるように精度の改善を目指す． | |||||||
| 蔀 竜太 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 | 加藤 博一 | 舩冨 卓哉 | 田中 賢一郎 | |
|
title: Direct Measurement of Wavelength Derivative of Spectra Using an Event Camera.
abstract: Spectroscopic measurement, in which decomposes light into each wavelength, is used in various applications such as component analysis of chemicals and quality control of foods. Particularly in the field of chemometrics, the wavelength derivative, which is the derivative of the spectral distribution with wavelength, is used to remove the effects of various noises.In the conventional method, the wavelength derivative is obtained by numerically differentiating the spectral distribution measured with a line sensor. However, in this method, the accuracy of the derivative depends on the wavelength resolution of the spectra and is greatly affected by the noise derived by the sensor.In this study, we propose a method of directly measuring the wavelength derivative independent from the wavelength resolution of the spectrometer. The proposed method uses an event camera that can directly measure changes in brightness values, and obtains the wavelength derivative by physically vibrating the event camera and recording the changes in adjacent spectral values. To prove the concept of proposed method, we build a spectroscope and confirmed that the event camera can record changes in spectral values. As a future work, we will obtain the wavelength derivative from the measured changes in the specula values. language of the presentation: Japanese 発表題目: イベントカメラを用いた分光の波長微分値の直接計測 発表概要: 光を波長ごとの成分に分解する分光計測は，薬品の成分分析や食品の品質管理などの様々な用途に用いられている．特に，計量化学の分野では，外乱による影響を取り除くために，分光分布を波長で微分した波長微分値が用いられる． 従来手法では，波長微分値はラインセンサを用いて計測された分光分布を数値微分して求める．ただし，この手法では微分の精度が分光の波長分解能に依存し，センサによるノイズの影響を大きく受けてしまう．本研究では，波長微分値を直接計測することで，分光の波長分解能に依存しない手法を提案する．提案手法では，輝度値の変化を直接計測できるイベントカメラを用い，イベントカメラを物理的に振動させて，隣接する分光値の変化を記録することで波長微分値を求める．提案手法を実現するために，分光器を製作し，イベントカメラで分光値の変化を記録できることを確認した．今後は，分光値の変化から波長微分値を求めていく． | |||||||
| ATUHURRA JESSE | M, 1回目発表 | 大規模システム管理 | 笠原 正治 | 門林 雄基 | 笹部 昌弘 | 張 元玉 | |
|
title: Online Intrusion-Detection for Internet-of-Things: A Machine-Learning-and-Fog-Computing Based Approach.
abstract: Due to the rapid increase in the number of IoT applications, it has become increasingly necessary to consider the security of such applications. Unfortunately,the existence of limited memory, power and computational capabilities makes the detection of attacks in IoT networks quite difficult. In this research, we propose an online intrusion detection system that harnesses the power of reinforcement learning to: learn the netwrok traffic patterns over time, and hence provide a more reliable security solutionfor the prediction and detection of attacks in an IoT application. language of the presentation: English | |||||||
| 菅原 琢哉 | 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 | 林 優一 | TRAN THI HONG | 張 任遠 | |
|
title: Co-optimization of Hardware and Software in Spiking Neural Networks abstract:In recent years, Artificial Neural Networks (ANNs) have been successful in many tasks. However, at the same time, high power consumption has become a problem. Therefore, Spiking Neural Networks (SNNs) have been attracting attention. SNNs are neural-inspired neural models that can be computed more efficiently than ANNs and are expected to be driven by low power consumption. However, low accuracy and difficulty in hardware implementation have been problems. In this paper, we propose an efficient implementation of SNNs in both hardware and software. language of the presentation: Japanese 発表題目: Spiking Neural Networkにおけるハードウェアとソフトウェアの両最適化 発表概要: 近年、Artificial Neural Neworks(ANNs)が多くのタスクで成果を挙げている。しかし、それと同時に高い消費電力が問題となっている。そこで、Spiking Neural Networks(SNNs) が注目されている。SNNsは神経インスパイアなNeural Modelで、ANNsよりも高効率に計算でき、低消費電力で駆動することが期待されている。しかし、低精度とハードウェアの実装しにくさが問題となっている。今回は、ハードウェアとソフトウェアの両面でSNNsを効率よく実装を提案する。 | |||||||
| TIAN HAOYU | 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: An Research For A Real-time Human Activity Recognition System Based On The Wearable Intelligent Device abstract: At present, the recognition of human activities based on sensors has become more and more important in personal healthcare. For example, a doctor can give a more effective treatment plan for the patient's biological signals. Although from the current point of view, some problems such as signal feature extraction, resource consumption when model building and other problems have been solved or improved. However, the real-time and accurate monitoring system for ultra-long time (24 hours or more) is still lacking. We proposed a prototype of a real-time monitoring system based on a single sensor unit of the wrist, and listed the current problems and corresponding solutions. language of the presentation: English | |||||||
| 奥村 嶺 | M, 1回目発表 | 情報基盤システム学 | 藤川 和利 | 安本 慶一 | 新井 イスマイル | ||
|
title: Indoor Positioning System Using Geomagnetism in a Garbage Incineration Plant
abstract: With the evolution of equipment in waste incineration plants, it has become possible to operate them with only a few people during normal times. However, the inspection and cleaning of the equipment are still dependent on workers. To ensure the safety of the workers, two or more people are required to work on these tasks, which is a hindrance to reducing personnel costs. Therefore, we thought it would be useful to have a system to share location information in real-time so that even a single worker can carry out his work safely. We will conduct a field survey and analyze the acquired data on indoor positioning using geomagnetism, and propose a system that is useful for waste incineration plants in terms of both implementation cost and effectiveness. Currently, we are in the stage of analyzing the acquired data after the first field survey. In the future, we will build a system that considers actual operation, and then confirm its effectiveness in different environments. language of the presentation: Japanese 発表題目: ごみ焼却施設における地磁気を用いた屋内測位システム 発表概要: ごみ焼却施設では機器の進化により、数人程度での平常時運用が可能となってきている。しかし、機器の点検や清掃は依然として、作業員に頼らざるを得ない。作業員の安全を考慮して2人以上で作業に当たる必要があり、人員コスト削減の足枷となっている。そこで、作業員が1人でも安全に業務を遂行できるよう、リアルタイムに位置情報を共有するシステムが有用ではと考えた。本研究では、地磁気を用いた屋内測位について現地調査と取得データ分析を行い、ごみ焼却施設において導入コストと効果の両方から有用なシステムを提案する。現在では、初回の現地調査を経て、取得したデータの分析に入る段階である。今後は、実際の運用も考慮したシステムの構築を行った後、異なる環境での有効性を確認する。 | |||||||