| 村重 哲史 | M, 1回目発表 | 数理情報学 | 池田 和司 |
|
title: Body Parts Tracking for Dogs towards Automatic Quantification of Dog’s Behaviors
abstract: In the research field of animal behaviors, video is one of the approach to record behaviors of animals. Even nowadays, however, recorded video data are analyzed with visual inspection and manual annotation, which are prohibitively inefficient for large scale data. It has been desired to develop an automated quantification method for behaviors of animals. We aim to develop a method that can track body parts of a dog, which can interact with humans as one of the social animals, in a video using computer vision methods. In this presentation, I will review the literature of body parts detection for animals including humans. In the literature, no method has been proposed for dogs. Also, I will report a preliminary result of single-dog detection using DeepLabCut, one of the deep learning techniques, with a video demonstration. language of the presentation: Japanese | |||
| 山本 明日翔 | M, 1回目発表 | 数理情報学 | 池田 和司 |
|
title: Elucidation of brain algorithm in learning conceptual knowledge by DecNef
abstract:In recent years, Convolutional Neural Network (CNN) which is the machine learning techniques imitating algorithm in the visual cortex of the human brain has attracted attention. The development of CNN makes it possible to classify of objects by image recognition. However, CNN has the problem that it requires a huge sample data for learning. On the other hand, human brain can learn concepts from small amount of sample data. For this reason, I aim to elucidate the brain learning algorithm for resolving the problem of CNN. The difference between CNN and visual cortex related to the number of sample data is in the feature extraction method. CNN extracts and learns all features from the entire image, while visual cortex extracts only necessary features with reference to knowledge own. For example, we can identify a house of any shape as a house by recognizing the door and the roof, the window. Thus, I hypothesized that it is possible to learn from a small amount of sample data by extracting and integrating some features as knowledge in the visual cortex and inputting them into the brain region involved in learning. In this colloquium, I will describe the tasks that is made to validate the hypothesis experiment subjects can learn conceptual knowledge and are useful in this study. language of the presentation:Japanese | |||
| 吉井 誉揮 | M, 1回目発表 | 数理情報学 | 池田 和司 |
|
title:Investigation of reward-driven learning mechanism by multivariate analysis
abstract:In recent years, Artificial intelligence(AI) has become showing performance beyond human ability in part. The background of the advance of AI has interaction with neuroscience. AI have made advance but there are still problems. Among other things, efficient learning and generalization are urgent problems. In the meta-RL of new framework, agent could learn quickly adapt in a variety of tasks a different rules. It is suggested that interaction between BG and PFC plays a role similar to this framework. We investigate interaction between BG and PFC by multivariate analysis. language of the presentation: Japanese | |||
| 森 純平 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 |
|
title: Sample value optimal estimation of rating transition matrix
abstract: In estimating future assets of investors and financial institutions, a rating transition matrix is used, but there is a problem that it can not be estimated that the classification transition matrix is not a specific cycle. In this research, we aim to develop a method to continuously complement the rating transition matrix by applying the idea of sample value control theory used for sound quality improvement of audio equipment. language of the presentation: Japanese | |||
| 安見 嘉人 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子 |
|
title: A Population Protocol for Uniform k-partition
abstract: In this research, we consider protocols that solve uniform k-partition probrem in a population protocol model. The goal of the uniform k-partition problem is to divide a population into k groups of the same size. As a first step, we consider the case of k=2. In this case, we study the problem under various assumptions: 1) a population with or without a base station, 2) weak or global fairness, 3) symmetric or asymmetric protocols, and 4) designated or arbitrary initial states. As a result, we completely clarify constant-space solvability of the uniform 2-partition problem and, if solvable, propose space-optimal protocols. Moreover, we give a symmetric k-partition protocol with designated initial states under global fairness. The proposed protocol requires 3k-2 states for each agent. Since any protocol for the uniform k-partition problem requires Ω(k) states to indicate a group, the space complexity of the proposed protocol is asymptotically optimal. language of the presentation: Japanese | |||
| 森谷 友香 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 |
|
title: Musculoskeletal Segmentation of Metal Artifact-Reduced Maxillofacial CT Images
abstract: The segmentation of the musculoskeletal structures, e.g. masseter muscle, in maxillofacial computed tomography (CT) images is essential for the diagnosis and treatment planning of maxillofacial disorders. However, the presence of a metal prosthesis, such as a dental filling, leads to metal artifacts in the CT images that affect the segmentation accuracy. In this research, we developed a musculoskeletal segmentation approach applied to metal artifact-reduced CT images. The artifact was reduced by using a normalized metal artifact reduction (NMAR) method. Next, the segmentation was performed by using a convolutional neural network, i.e. two-dimensional (2D) U-Net. The proposed approach was tested on CT images of two patients. An average segmentation error (average symmetric surface distance (ASD)) of 2.6 mm was obtained for the segmentation of the masseter muscle. Future work will focus on applying the proposed method to larger datasets, and the segmentation of another maxillofacial muscles. language of the presentation: Japanese | |||
| 八木 聖太 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 |
|
title: Transmission power control for multiple terminal connections in wireless communication
abstract: To perform multiple simultaneous connections of terminals in a wireless network, stable transmission power control is required. The Foschini-Miljanic algorithm, which has been widely used for transmission power control, may have reduced stability due to parameter uncertainty. In this research, we stabilize algorithms by adjusting terminal specific parameters. language of the presentation: Japanese 発表題目: 無線通信における端末の多数接続のための送信電力制御 発表概要: 無線ネットワークにおいて端末の多数同時接続を行うには，安定的な送信電力制御が必要となる．送信電力制御に幅広く用いられてきたFoschini-Miljanicアルゴリズムは，パラメータの不確実性によって安定性が低下する可能性がある．本研究では，端末固有のパラメータを調整することでアルゴリズムの安定化を行う． | |||
| 恒松 和輝 | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title: Listen to the Hint, and Speak with Your Own Words
abstract: In human-machine interaction, recent technologies such as speech recognition (ASR) and speech synthesis (TTS) have enabled a machine to listen and speak like a human. But ASR only passively transcribes what was being said. TTS only passively read out loud what was being written. On the other hand, human often listens to incomplete sentence and being able to generate the complete sentence using their own words. In this research, we aim at constructing a machine which can perform such task using deep learning. language of the presentation: Japanese | |||
| 山﨑 勇二 | M, 1回目発表 | 情報基盤システム学 | 藤川 和利 |
|
title: Code-Reuse Attacks for the Web: Breaking Cross-Site Scripting Mitigations via Script Gadgets
abstract: Cross-Site Scripting (XSS) is an unremitting problem for the Web. Because of this, various approaches to mitigate XSS have been proposed as a second line of defense, with HTML sanitizers, Web Application Firewalls, browser-based XSS filters, and the Content Security Policy being some prominent examples. In this presentation, we demonstrate that this approach is no longer sufficient for modern applications: We describe a novel Web attack that can circumvent all of theses currently existing XSS mitigation techniques. In this attack, the attacker abuses so called script gadgets (legitimate JavaScript fragments within an application’s legitimate code base) to execute JavaScript. After that, we will discuss the consideration of DOM Based XSS prevention method. language of the presentation: Japanese | |||
| 豊 美玲 | M, 1回目発表 | 大規模システム管理 | 笠原 正治 |
|
title: Attribute-Based Access Control Using Block Chain Technology
abstract: In the Internet of Things (IoT), a huge number of devices (e.g., sensors and actuators) are deployed around us to gather sensitive information (e.g., health data) and conduct safety-critical tasks (e.g., car braking). As a result, illegal access to these devices poses significant threats to our personal safety. To prevent these illegal access, this research proposes a distributed and dynamic access control scheme for IoT systems based on the emerging blockchain technology and attribute-based access control model. language of the presentation: Japanese | |||
| 桃井 雄資 | M, 1回目発表 | 大規模システム管理 | 笠原 正治 |
|
title: On Enumeration of Stable Matching Solutions by Zero-suppressed Binary Decision Diagram
abstract: Making a good matching between two-layer vertices is one of the most important problems not only in mathematics but also in practical situations. There are a huge number of candidates for matching solutions in a graph. Each vertex has its own wish list of vertices in the other side. If vertices are matched to dislikes or they feel they could get more appropriate partner, they may complain of their matching result. Stable matching is one of the solutions for their discontent, because no other pair likes each other more than matching pairs. In this reaserch, we consider to enumerate all possible stable matchings by using ZDD, which is a data structure for representing a set family, in order to offer good choices in large scale matching problems. language of the presentation: Japanese | |||
| 山本 将成 | M, 1回目発表 | 大規模システム管理 | 笠原 正治 |
|
title: On Evaluation of Block Distribution Time in Bitcoin Network with Malicious Node
abstract: Bitcoin is a cryptocurrency developed based on the paper written by Satoshi Nakamoto. The Bitcoin system relies on an unstructured Peer-to-Peer(P2P) network. Although nodes are expected to propagate information through the network according to the regular Bitcoin networking protocol, some malicious nodes can cheat the protocol by modifying the original source code. As a result, the risk of attack using timeout at block transfer has been reported. In this study, we evaluate how block distribution time will be delayed by the malicious nodes using the delay attack, through simulation experiments. language of the presentation: Japanese | |||
| 山元 勇輝 | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Abstract Meaning Representation Parsing: Lexical-Semantic Approach
abstract: Semantic Parsing is a task which converts sentences into semantic representations. Recently, a sentence-level semantic formalism called Abstract Meaning Representation (AMR) has been proposed. Various approaches have been taken to parse sentences into AMR formalism. In this presentation, we show some related works and propose a method from the standpoint of lexical semantics. language of the presentation: Japanese | |||
| 由井 朋子 | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Technical Quantification and Evaluation of Dental Hygienists in Hand Scaling
abstract: Dental hygienist students need to learn how to perform dental scaling by hand scalers. However, this is not easy to learn. Therefore, if this work can be evaluated quantitatively, we can standardize this technology, develop self-study tools so that it will be helpful for education. In this research, we propose a method to measure gaze and finger movements when a dental hygienist is performing scaling with a hand scaler, also attempt to quantify and evaluate this technology. language of the presentation: Japanese 発表題目: ハンドスケーリングにおける歯科衛生士の技術定量化とその評価 発表概要: 歯科衛生士学生は、ハンドスケーラーによって行うスケーリング作業を学ぶ。ただ、この作業は容易に習得できる作業ではない。そこで、この作業を定量的に評価することが可能になると、その情報から技術の標準化や自習ツール開発が可能となり教育に活用できる。 よって本研究では、歯科衛生士がハンドスケーラーにてスケーリングを行う際の視線や指の動きを測定する方法を提案し、技術の定量化と評価を試み検討する。 | |||
| 山田 憲 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 |
|
title:Development of structured light system for measuring metallic objects
abstract: Techniques for measuring three-dimensional shapes of metallic objects are demanding in industrial area because metals are used in various industrial products. However, metals have strong interreflections so it is difficult to perform stable measuring. In this research we propose a method to realize the structured light system to capture three-dimensional shape of metallic object with strong interreflection. language of the presentation:japanese 発表題目:金属計測のための構造化光システムの開発 発表概要: 金属は様々な工業製品に使用されており，それらの三次元形状を計測する技術には大きな需要がある． 一方で金属は強力な相互反射を持つため計測を行うのは困難である． そこで本研究では相互反射がある環境でも頑健に取れる構造化光システムの開発を目的とし， 金属に対しても安定した形状計測が可能な構造化光システムを実現させるための第一ステップを紹介する． | |||
| 森島 僚平 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 |
|
title: Automating interpolation of frames for making animations using FCNN
abstract: Japanese animation becomes more popular around the world as one of the Japanese cultures. However, the process for making the animations are complicated and time consuming. In order to make animations more efficiently, we decided to use the deep learning, which has greatly developed various research related to computer graphics in recent years. In this research, we propose a method using FCNN to realize the frame interpolation automatically, which is important process of an animation production. language of the presentation: Japanese 発表題目: FCNNによるアニメーション制作におけるフレーム補間の自動化 発表概要: 日本のアニメ産業は世界に誇ることができる日本の文化の一つである。しかし、アニメ制作の工程は大変複雑である為、制作にかかるコストは大きい。我々はこのアニメ制作の作業を効率化させるのに、ここ数年でコンピュータグラフィックスに関わる様々な研究を大幅に発展させてきたディープラーニングが利用できると考えている。そこで本研究で我々はアニメ制作の工程の一部であるフレーム補間をFCNNを用いて自動化する手法に取り組んだ。 | |||
| 吉田 茉菜 | M, 1回目発表 | モバイルコンピューティング | 伊藤 実 |
|
title: Consideration of user cancellation in pre-order road pricing system
abstract: Traffic congestion is a serious problem that causes significant social losses. Especially, expressways enable users to quickly reach their destinations by paying fees, but the traffic congestion is one of the main factors to strongly disturb the advantage. Various techniques have been studied to reduce traffic congestion. One of them is real-time road pricing on expressways. In particular, in Japan, there are many users who use Electronic Toll Collection System (ETC), the real-time road pricing is expected on expressways and to cooperate with pre-order systems in the future. In order to realize real-time road pricing with pre-order systems, the price determination method is important to efficiently reduce the road congestion and needs to consider updates of the pre-ordered information. In this research, we will study the price determination method considering user cancellation on expressways for pre-order systems. language of the presentation: Japanese | |||
| 和田 遥香 | 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 of ibeacon-based Mobile Application with View Switching and Information Supplement Functions Depending on Current Position of Users
abstract: Making nursing report by workers is mandatory by law and is important for sharing information and improving services. However, it becomes a burden of workers. And it is said that 25% of nursing care time is spared, and the time to do essential nursing care work is decreasing. We conducted a system to automatically generate care reports by using BLE beacons in previous research. However, in this study, only records of the time-series rough behavior were recorded. Therefore, it is necessary to take detailed records including the location information (e.g. excretion in the toilet, the meal at the dining,etc.) In this research, we propose a mobile memo system which automatically switches the information input screen and its items depending on the location, specifies the user and complements the information. As future work, we will experiment and evaluate this system in a real daycare center and let people in there use this system. language of the presentation: Japanese | |||