| 柏本 幸俊 | 1251029: M, 1回目発表 | ユビキタスコンピューティングシステム |
|
title: Implementation and Evaluation of "Wisdom of Life" Sensing
toward Energy Saving Activity Support
abstract: After the 2011 Tohoku Earthquake, Japan faces severe power supply shortage and people are required to maintain spontaneous efforts in saving electric energy consumption. In this research, we aim to support users to save electric energy consumption without too much deterioration in users’ comfort perception by utilizing potential wisdoms of daily lives. The proposed system extracts wisdoms of daily living that contribute electric energy saving from users who are dedicated to eco-behavior using the electric power meters attached to appliances, environmental sensors, and behavioral sensors. To extract wisdoms of daily lives, it is necessary to find relationship between users’ behavior and comfort level in a certain time period, and to find behavior patterns that achieve both low electric energy consumption and high comfort perception. Therefore, our system extracts energy-efficient and comfortable behaviors from multiple users who have the similar living pattern by analysing sensor data stored in databases and by comparing these users’ behaviors. The extracted wisdoms are provided with display devices to support users who are not good at eco-behavior. Using sensors described above, we demonstrate our system, which detects wisdoms of daily lives such as water sprinkling for cooling. language of the presentation: Japanese 発表題目: 省エネ行動支援に向けた"生活の知恵"センシングシステムの実装と評価 発表概要: 東日本大震災以降電力需給が逼迫しており，各電力ユーザに自主的な節電が求められている．本研究では，日常生活に潜在する生活の知恵を活用することで，ユーザの快適度を下げずに節電を支援することを目的とする． 提案手法では，各家電に取り付けられた電力計，環境センサやユーザの行動センサのデータをもとに，節電が上手なユーザの行動に潜在する生活の知恵を抽出し，節電が不得手なユーザを支援する． 生活の知恵を抽出するには，ある時間間隔におけるユーザの行動と快適度，電力消費を紐づけ，その上で，低消費電力かつ快適性の高い行動パターンを見つける必要がある． そこで本手法では，複数のセンサデータを蓄積・解析し，生活パターンの類似する複数のユーザの行動を比較し，その中で節電効果が高くかつユーザの快適度を低下させない行動，すなわち潜在的な生活の知恵を抽出し，節電が不得手なユーザに，表示デバイスを介しその知恵を提示することで支援を行う． | ||
| 藤原 知広 | 1251091: M, 1回目発表 | コンピューティング・アーキテクチャ |
| title: Research on effective prefetch of accelerator abstract: As is necessary to improve the memory latency, faster GPGPU need to improve the latency problem between host / accelerator accelerator even faster. This time I presented the background of the study and related research through the introduction of references that describe an adaptive prefetch throttling scheme, which permits automatic GPGPU application- and hardware-specic adjustment. language of the presentation: Japanese | ||
| 東 健太郎 | 1251005: M, 1回目発表 | インターネット工学 |
| title: deliberation about web on P2P abstract: Recently, cyberattack is widely known and will be very serious problem from now on. Many organizations have paid attention to network security, but there is no silver bullet. In this presentation, I try to web which is robust over the attacks by substituting web on P2P for conventional web. I discuss about problems of my proposals. language of the presentation: Japanese 発表題目: P2Pを利用したweb実現の検討 発表概要: 近年、広く知られるようになったサイバー攻撃は今後無視できない存在になると考えられる。各組織ではネットワークセキュリティの強化を図っているが、決定的な対策は存在していない。本発表では、これらの攻撃に強いwebの実現を目指し、コンテンツを周囲に公開されたサーバに配置する従来のwebからP2Pネットワーク上に配置するwebを目指す。その上で考えられる問題点について参考文献を挙げながら、P2Pネットワークを用いたwebの実現について検討する。 | ||
| 森 健太 | 1251107: M, 1回目発表 | 情報基盤システム学 |
|
title: Congestion Control by Harmonized Channel Selection for Intelligent Transport Systems
abstract: Intelligent Transportation Systems (ITS) is a generic name of system groups that contribute to the improvements in transport efficiency, comfort and safety using information technology. Specifically, communication system of vehicle-to-vehicle and vehicle-to-infrastructure for safety using IEEE 802.11p has been promoted standardization. However, owing to the nature of contention based channel access scheme, the system suffers from Quality of Service (QoS) degradation for safety applications caused by the channel congestion in scenarios with the heavy traffic road. Therefore, various congestion control approaches are proposed. In this presentation, I introduce several congestion control approaches. Furthermore, I also talk about a primitive idea of my research. language of the presentation: Japanese | ||
| 丹生 伊左夫 | 1251079: M, 1回目発表 | 知能コミュニケーション |
|
title : Extracting Most Minimum Training Data For Statistical Machine Translation
abstract : Since internationalization of society is annually progressing, the occaion that many people who speak various languages have a conversation is increasing. However spoken languages vary from person to person, it is difficult to smoothly communicate each other. Then to translate one language to other languages, Machine Translation System is researched actively. However the accuracy of translation is not high enough to use practical case. There is Statistical Machine Translation as one method of the Machine Translation. This method is that statistically translate one language to other languages. In Statistical Machine Translation, the data for learning this system affect the accuracy of translation. From the above, I introduce the method that automatically extract this training data from Web for Statistical Machine Translation. And my destination is that I find most minimum training data enough to use practical use. language of the presentation : Japanese | ||
| 長岡 博 | 1251074: M, 1回目発表 | 視覚情報メディア |
|
title: Paper introduction "DEPTH-BASED IMAGE COMPLETION FOR VIEW SYNTHESIS" abstract: This presentation is paper introduction of "DEPTH-BASED IMAGE COMPLETION FOR VIEW SYNTHESIS". I'm going to research about image inpainting with depth image. This paper proposes a depth based inpainting algorithm for vertual viewport rendering. Using the relationship between the position of vertual viewpoint and target regions of restration this algorithm can create clearer restored images than other methods. language of the presentation: Japanese | ||
| 重藤 優太郎 | 1251050: M, 1回目発表 | 自然言語処理学 |
|
title: Construction of English Multiword Dictionary and its Application to POS Tagging
abstract: Many previous studies have proposed various methods for English Part-Of-Speech (POS) tagging, and these methods have been shown to give high accuracy. However POS tagging in these previous methods are only on each word, and they cannot tag on multiword expressions consisting of multiple lexical items, such as collocations. In this research we constructed a multiword expression dictionary, and we used variable length CRF to conduct POS tagging which takes into consideration the multiword expressions. We, then, evaluated the tagging accuracy and the multiword recognition accuracy. In result, the tagger with the multiword dictionary achieved higher accuracy compared with the tagger which does not consider multiword expressions. language of the presentation: Japanese | ||
| 北村 大地 | 1251035: M, 1回目発表 | 音情報処理学 |
|
title: Musical Audio Source Separation by Penalized Supervised Non-Negative Matrix Factorization for Actual Instrumental Sound
abstract: For the source separation of musical audio signal that consists of multiple instrumental sources, Non-Negative Matrix Factorization (NMF) is often used. NMF decomposes spectrogram into frequently-appearing spectral pattern and its activation. However, It is problem that decomposed spectral pattern contains multiple instrumental spectra. To solve this problem, Supervised NMF that utilizes spectrum of the target instrumental MIDI sound as supervision information was proposed. Nevertheless the distinction between supervision sound and the target actual sound reduces the accuracy of separation. In this study, we seek to improve the proposed method to separate the actual sound utilizing MIDI supervision sound. language of the presentation: Japanese 発表題目: 実楽器音を対象とした罰則条件付き教師あり非負値行列因子分解による音楽音源分離 発表概要: 種々の楽器音が多重に録音された音楽音響信号から目的の楽器音を抽出する音楽音源分離技術には，よく非負値行列因子分解 (NMF) が用いられる．NMFは，音響音楽信号のスペクトログラムを頻出するスペクトルパターン (基底) とそれらの時間的な音量変化 (アクティベーション) に分解する手法であるが，分解した頻出スペクトルに複数の楽器音が混在してしまうという問題がある．この問題に対し，あらかじめ抽出したい楽器のMIDI音のスペクトルを教師情報として与えて分解する教師ありNMFが提案された．しかし，実楽器音を対象とした教師ありNMFでは，抽出目的である実楽器音とMIDI教師音の音色の差異により抽出精度が低下してしまう．本研究では，従来の教師ありNMFに対して分解条件と罰則条件の改良を加え，MIDI教師音を用いた実楽器音源分離を目指す． | ||
| 木谷 竜也 | 1251034: M, 1回目発表 | ネットワークシステム学 |
|
title: Base station selection for distribution antenna system
abstract: Distributed antenna system (DAS) is that multiple antennas distributed are connected by optical fiber socalled RoF (Radio over Fiber) link and each fiber connected central control station(CCS). CCS performs signal proccesing and controls scheduling(time,frequency), base station selection. This presentation focuses on base station selection. Problem in base station selection is that how to select base station to perform efficient transmission. This time, I present proposal method , its result and future task. language of the presentation: Japanese 発表題目: 分散アンテナシステムにおける基地局選択法 発表概要: 分散アンテナシステムとはあるエリア内に多数の基地局を分散配置しそれぞれを光ファイバで繋ぎ中央制御局に接続する方法である.中央制御局では信号処理,スケジューリング（時間,周波数）,基地局選択などの処理を一括して行う.今回は基地局選択に焦点を絞って発表する.基地局選択において課題となってくるのが効率的なMIMO伝送を行うためにどのように基地局を選択するかということである.本発表では,提案手法,提案方式による結果,今後の課題について述べる. | ||
| MATEO,LORLYNN ASUNCION | 1251127: M, 1回目発表 | 知能システム制御 |
|
Title: Optimal Sensor Scheduling in Wireless Sensor Networks
Abstract: In order to provide a reliable coverage when collecting physical data from the environment, different resource allocation techniques must be applied to wireless sensor networks (WSNs). Restrictions in communication bandwidth, memory and on-board power may inhibit a WSN from delivering its maximum performance. One technique is sensor scheduling, where sensing resources are allocated by optimizing a performance metric under a given set of constraints over a future time horizon. From the perspective of Networked Control Systems, different implementations of sensor scheduling were surveyed. A specific example using a Kalman Filter-based state estimator, which is based on the paper entitled “Optimal Periodic Sensor Scheduling with Limited Resources”, shows concrete steps on how to model a sensor scheduling problem as a discrete linear time-invariant system. The paper then proposes a method to construct an optimal periodic schedule where only one sensor can send measurements per time step. Language of the presentation: English | ||
| 伊藤 峻大郎 | 1251011: M, 1回目発表 | 生命機能計測学 |
|
title:Development of a support tool for creating animations protein complex abstract:Protein in the human body is subjected to enzymatic reaction becomes complex. Then, the state of the reaction has been elucidated by theoretical calculations, such as molecular dynamics. However, since it is difficult to visually confirm the fact, it is difficult to imagine that the enzymatic reaction. Therefore, there is a growing need to enzymatic reactions Film. In this study, we develop a software that can easily create and edit videos molecule. In this presentation, Reveals the problems of the existing software, we describe the measures and plans for future research. Language of the presentation:Japanese 発表題目:タンパク質複合体アニメーションの作成支援ツールの開発 発表概要:人間の体内ではタンパク質は複合体となって酵素反応を行っており, 反応の様子は分子動力学等の理論計算によって明らかにされている. しかし,実際に目視の確認は困難であり、酵素反応をイメージする事は難しい. そこで,計算結果を動画にする必要性が高まってきている. 本研究では,簡単に動画の作成や編集が可能なソフトウェアを開発する. 本発表では,既存のソフトウェアの問題点を提起し,対策と今後の研究予定について述べる. | ||
| 豊吉 政彦 | 1251070: M, 1回目発表 | ロボティクス |
|
title: Random Forest for Multi-class object detection and pose estimation using RGB-D camera
abstract: Fast and reliable algorithms for Multi-class object detection and pose estimation are important issue in robotics and computer vision research. We propose an efficient multi-class object detection and pose estimation approach in large-scale real-world. Random forests have been shown to handle large training datasets and many classes for object detection efficiently. And previous work shows Random Forest is fast and reliable algorithm for head pose estimation. We address the problem of multi-class object detection and pose estimation from RGB-D data, which can be captured using the ever more affordable RGB-D sensing technologies available today. In this presentation, We introduce the previous works include an explanation of Random Forest and future development. language of the presentation: Japanese 発表題目: 多クラス物体の検出と位置・姿勢推定のためのRGB-Dセンサーを用いたRandom Forestアルゴリズム 発表概要: ロボティクスやコンピュータビジョンの分野において，物体を検出し，その位置と姿勢を推定する事は，物体の把持等のために，重要な技術である．本研究では，大きなスケールの実世界において，様々な種類の物体を検出し，その物体の位置と姿勢を推定する，効率的な手法を提案する．Random Forestアルゴリズムは，画像処理の分野においてこれまでに，多くの物体の種類を含む大量の訓練データセットを用いて物体検出をする際の，優れた手法である事が示されてきた．また，Depthセンサーを用いたRandom Forestアルゴリズムによって，人間の頭部の位置・姿勢の推定をした先行研究等においても，効率的な手法であるという結果が出ている．本研究では，これらの結果を受けて，通常のカメラとdepthセンサーを搭載したRGB-Dセンサーを用いて，Random Forestアルゴリズムによって，物体を検出し，さらにその位置・姿勢を推定する事を目的とする．本発表では，Random Forestアルゴリズムの解説を含む，先行研究についての紹介と，今後の研究の展開について述べる． | ||
| 吉澤 知彦 | 1251116: M, 1回目発表 | 神経計算学 |
|
title: A study of hierarchical brain representations of model-based strategy in decision making abstract: Decision making is a mental process to choose one of the multiple choices. The cortico-basal ganglia loops are recognized as the major sites for decision making and reinforcement learning. Model-based strategy has been proposed as a model of a goal-directed decision making, but little is known about how model-based strategy is represented in the brain. On the other hand, it has been proposed that the cortico-basal ganglia loops including the dorsolateral, the dorsomedial, or the ventral striatum comprise a parallel and hierarchical reinforcement learning modules that are in charge of actions at different physical and temporal scales. But there is no direct evidence for this hypothesis yet. An overarching goal of the project is to elucidate the role of the cortico-basal ganglia loops in model-based decision making. The research questions to be addressed are: 1) Are there the signals related to model-based strategy in the cortico-basal ganglia loops? 2) Are the high-level decision making and the low-level decision making processed simultaneously in the cortico-basal ganglia loops? For these purposes, rats will be trained to perform a model-based choice task, in which an action selection using the knowledge of an association between a tone and a reward is required. While they perform the task, their head position and body position will be recorded by 3D tracking system, and neuronal activity in the dorsomedial, and the dorsolateral striatum will be recorded by the extracellular recording method. Then, we will analyze the neuronal activity to find representations related to model-based strategy and hierarchical decision making. language of the presentation: Japanese | ||