| 木村 歩賛名 | 1251038: M, 1回目発表 | インタラクティブメディア設計学 |
|
title: Development of the System Growing in Capability of Reading a Map by Supporting Aberration Detection
abstract: In recent years, the navigation system has permeated our life rapidly. But, we can't use the system when some sensors such as GPS, electronic compass don't work accurately in the city or the terminal what provided navigation is out of charge. In this case, the user who have trouble reading a map and depend on the navigation system will involve taking risk of getting lost. This problem is attributable to the system's purpose that is not so much removal of the underlying cause but getting to the destination. So we focus attention on aberration detection technique used in orienteering and propose the system that solve the problem about getting lost. language of the presentation: Japanese 発表題目: 逸脱検出支援による地図読み技術向上システムの開発 発表概要: ナビゲーションシステム(以下ナビ)は近年になって急速に一般に浸透してきた. しかしGPSや電子コンパスなどのセンサが都市部に弱い問題や, ナビを利用するスマートフォンなどの端末が抱えるバッテリーの問題などによりナビを利用できない環境・状況は決して少なくない. このような状況において元々地図を読むことを苦手としてナビに依存していたユーザはたちまち道に迷うリスクを負うこととなる. この問題は現状のナビが, 目的地にたどり着くことのみを目的としており道に迷うことの根本的な解決を目指すものではないことに由来すると考えられる. そこで本研究では, 地図読みの競技, オリエンテーリングにおける技術の一つである『逸脱』の検出に着目して, 道迷い問題の根本的解決を図るシステムを提案する. | ||
| 松田 勝志 | 1251096: M, 1回目発表 | ネットワークシステム学 |
|
title: Thoracoscopic surgery support system using passive RFID marker
abstract:An RFID based thoracoscopic surgery support system, which is capable of marking a tumor inside organ tissue, has been proposed. In this system, first the marker composed of small RFID-tags is implanted in the vicinity of tumor found in the endoscopy test. Then, in the thoracoscopic surgery operation for removing the tumor, an RFID detector determines the accurate position of the implanted RFID-tag markers by measuring the strength of the signal emitted from the target tag. Due to limitation in the size of RFID-tag, this system employs a passive RFID. However, this system sometimes can't determine the position of the implanted RFID-tag markers because of the angle between a tag and a sensor antenna, so I propose a scheme which uses multiple sensor antennas to solve this problem. language of the presentation: Japanese | ||
| 赤熊 高行 | 1251002: M, 1回目発表 | 視覚情報メディア |
|
title: Study on Geometric Consistency in Mobile Augmented Reality
abstract: Recently, a wide variety of applications have been proposed for augmented reality systems for mobile devices (mobile AR systems) such as providing tourist information. Large difference between a camera position and a user’s viewpoint in mobile AR raises two problems of geometric consistency, i.e., (��) real-time alignment of virtual objects in a real scene, which is a conventional problem in the field of augmented reality, as well as (��) a gap between a camera position and a user’s viewpoint, which may affect the performance of mobile AR systems such as its usability and naturalness. In this study, we propose a high-speed geometric registration method for (��)', which focuses on a biased distribution of feature points, and verify effect of the gap between a camera position and a user’s viewpoint by a user study. This presentation describes the progress of this work. language of the presentation: Japanese 発表題目: モバイル端末における拡張現実感のための幾何学的整合性に関する研究 発表概要: モバイル端末上で仮想物体を重畳した映像を表示する拡張現実感（Augmented Reality; AR）システムは，観光案内をはじめとした様々な応用が提案されている．モバイル端末上でのARシステムは，カメラの位置とユーザの視点が大きく異なるため，(��) 仮想物体と実シーンの幾何学的整合性の実時間での解決に加え，(��) ユーザがカメラを通して見る実シーンとユーザ視点からの実シーンでズレが生じ，ARシステムに影響を与える可能性がある．これら２つの幾何学的整合性の問題があるため，本研究では，(��)' 特徴点の分布の偏りに着目した高速な位置合わせ手法の提案、及び(��)' カメラの位置とユーザの視点による幾何学的不整合の影響をユーザスタディによって検証する．本発表では，現在の進捗を報告する． | ||
| 岡田 亜沙美 | 1251023: M, 1回目発表 | 環境知能学 |
|
title: People Groping by Spatio-Temporal Features of Trajectories abstract: We proposes a method for detecting people groups from their trajectory data. This grouping is applied to each pair of people. The trajectories of the pair are featured by their spatio-temporal relationships such as a distance and velocities. The features are classified to either of “group” or “non-group” by a discriminative classifier. In contrast to previous features, the proposed features are robust to unsteady behaviors of people and noise of their trajectories. Experimental results using a publicly-available dataset of trajectories demonstrate the effectiveness of the proposed method. language of the presentation: Japanese | ||
| Frances Pikyu Yung | 1251126: M, 1回目発表 | 自然言語処理学 |
|
title: Learning Core Word Alignments for Statistical Machine Translation
abstract: Statistical machine translation systems learn translation rules from human translated sentences aligned to the word level. Typically, word alignments are carried out automatically basing on the how often each bilingual word pair occurs. When training data is limited, word alignments inferred by co-occurance statistics may undesirably include frequent grammatical words. It is also intuitively inappropriate to include contextually inferred words as core lexical translations, yet it is a usual practice for human translators to insert these words. This 'attached word' phenomenon is particularly common for translation between distant languages where many words do not have direct correspondence in the translation. In this presentation, I will describe our hypothesis, supported by preliminary experimental results, that removing attached words in the training data helps improve word alignments and acquiring these words at a later stage can improve statistical machine translation. language of the presentation: English | ||
| 田中 宏 | 1251058: M, 1回目発表 | 知能コミュニケーション |
|
title: electricity speech enhancement based on Noise suppression and Statistical speech production
abstract: Laryngectomees need some alternative methods of speaking to produce alaryngeal speech. Their common problem is that any type of alaryngeal speech suffers from lack of naturalness, speaker individuality and clearly. To address this problem, we have one approach. That is a hybrid method between spectrum subtraction and statistical voice conversion. In this presentation, I show you details of these methods. language of the presentation: Japanese | ||
| 宮内 智 | 1251103: M, 1回目発表 | 音情報処理学 |
|
title: Perceptual Evaluation of Synthesized Sound Field with Separated Signals from Stereo Music Mixtures
abstract: Wave Field Synthesis (WFS) enables to reproduce highly realistic sounds. However, in conventional contents, primary source information needed in WFS has been degenerated. For this reason, the conventional contents cannot be apply to WFS. Therefore, automatic estimation method to recover spatial information of the primary source from conventional content is required. In this presentation, as a preliminary step of automatic estimation method, we report the results of a subjective evaluation experiment of distance perception in synthesized sound field using source separation based on audio objects and WFS. This evaluation was carried out to consider the problem of the possibility of distance perception in the sound field synthesis using separated signals from stereo music mixtures. language of the presentation: Japanese | ||
| 和良品 友大 | 1251121: M, 1回目発表 | 自然言語処理学 |
|
title: A Fast Approximate Simirarity Search with k-Nearest Neighbor Graph
abstract: A similarity search that finds valid objects similar to a given query object has been a fundamental technique for many applications. Some applications require to obtain a solution quickly, even if it is not always exact, rather than an exact solution after a long computation time, particularly when dealing wiht large-scale and high dimensional data. I will describe a fast approximate simirarity search method based on k-nearest neighbor graph for large-scale and high dimensinal data. A k-nearest neighbor graph has small-world properties, which is the tendency of an extremely small average shotest path length. Because of this property, our method has high search performance. However, the graph is difficult to construct with parallel processing. I will also introduce an alternative method. The method utilizes a graph similar to k-nearest neighbor graph, which is easy to construct with parallel processing. language of the presentation: Japanese | ||
| 清水 宏晃 | 1251051: M, 1回目発表 | 知能コミュニケーション |
|
title: Hierarchical phrase-based model for real-time speech translation system
abstract: In recent years, there has been increasing to communicate across languages barriers. And the demand for speech translation system that can automatically translate a language into another language has also been increasing. However, when the speech translation system is used, there is a speech delay that is from the beginning of the utterance to the beginning of the speech synthesis. To shorten the speech delay can realize to communicate in real-time. So, it is difficult to communicate using the speech translation system, because a user have to wait during the delay. In this presentation, I proposed a real-time speech translation system that shortens the speech delay with keeping the translation accuracy. language of the presentation: Japanese | ||
| 藤原 雄介 | 1251090: M, 1回目発表 | ソフトウェア設計学 |
|
title: Support for Modification of Code Clone Cooperated with Version Control System
abstract: Code clones are identified as the sections of code (clone pieces) which are duplicated or similar to each other in a source code. It is generally believed that code clones prevent software developers to perform software maintenance tasks adequately. For example, when any of the clone pieces contained in code clone is modified, all other clone pieces need to be taken into consideration whether they should be consistently modified or not. If such consistent modifications are missed, they could become the cause of future bugs. There exist many methods and tools which detect code clones. However, in order to check consistent modifications, those tools need to be launched every time the source code is modified, making their use an extra burden for software developers. In this study, we propose an approach that automatically detects inconsistent modifications of code clones. So that, when a code clone is modified and committed to version control systems, inconsistent modifications are automatically noticed. language of the presentation: Japanese 発表題目: バージョン管理システムと連携したコードクローンの一貫した修正支援 | ||
| Chu Bao Trung | 1251123: M, 1回目発表 | 計算メカニズム学 |
|
title: An effective method for quantitative information flow analysis
abstract: Classical information flow analysis concerns non-interference, i.e., whether a high security level data never flows into a low security level output channel. However, non-interference is too restrictive notion for determining the secrecy of real-world programs. To overcome this disadvantage, researchers came up with quantitative information flow (QIF), which can be used to measure the amount of information flow through a program based on information theory. Though QIF is very useful, we still do not have an effective way to analyze it. In my talk, I would like to explain a new method proposed by Kopf & Rybalchenko, which uses approximation and randomization to effectively analyze the QIF. Also I will present my research plan to implement the method to analyze the QIF and to apply the method to automatic generation of access control. language of the presentation: Japanese 発表題目: 効率的な量的情報流解析法 発表概要: 古典的な情報流分析法では、非干渉性、すなわち、高い機密レベルのデータが低い機密レベルの出力に決して流れないかどうかを検査します。しかし実際に使われているプログラムにとって非干渉は制限が強すぎます。この問題点を解決するため、量的情報流が導入されました。量的情報流解析では、情報理論に基づきプログラムを流れる情報量を定量的に計算します。 量的情報流はこのように役に立つものの、現在はまだその効率的解析法は知られていません。今回の発表ではKopfとRybalchenkoが提案した新しい解析法を紹介します。これはプログラムの抽象化による近似、及びランダム化を使って量的情報流を解析する効率的な方法です。続いて発表者の研究計画についてもお話します。発表者は今回紹介した解析法を実装し、またその解析法をアクセス制御の自動的生成問題に応用する予定です。 | ||
| Yussuf Ali | 1251130: M, 1回目発表 | ディペンダブルシステム学 |
|
title: Scan Test Power Simulation on GPGPUs
abstract: During Scan-Test the chip encounters a much higher switching activity compared to the normal functional mode. This high switching activity results in higher power consumption. Much research is done in order to optimize scan testing. The behavior of the chip during scan testing can be simulated with the help of a timing simulator. Depending on the size of the chip this simulation process can take many hours or even days. To accelerate the simulation process, the GPU can be used. Because during the scan test all input pattern are known in advance, all patterns can be simulated at the same time. In the presentation I will introduce a timing simulator, which uses the GPU, one of the key features of this simulator is the support for hazards. language of the presentation: Japanese | ||
| 村田 大輔 | 1251106: M, 1回目発表 | 情報基盤システム学 |
|
title: Network security in a IPv4-IPv6 address mixed network
abstract: As new Internet Protocol version 6 (IPv6) emerged, many people believe this protocol can provide good protection against network worms due to its huge address space. However, IPv6 is still new compares to native IPv4, some transition mechanisms have been introduced to promote smoother migration. Unfortunately, these transition mechanisms allow some malwares to propagate its attack from IPv4 to IPv6 network environment. So, IPv4-IPv6 dual-stack network is more dangerous than pure IPv4 Internet. In this presentation, I will talk about malware behave differently on IPv6 from IPv4 network protocol on the dual-stack network and a lot of things that want to do. language of the presentation: Japanese | ||