| 竹森 有祐 | 1351065: M, 1回目発表 | 環境知能学 |
|
title: [Paper Introduction] Tourist Trip Planning Functionalities:State-of-the-Art and Future
abstract: When tourists visit a city or region, they cannot visit every point of interest available, as they are constrained in time and budget. Tourist recommender applications help tourists by presenting a personal selection. Providing adequate tour scheduling support for these kinds of applications is a daunting task for the application developer. The objective of this paper is to demonstrate how existing models from the field of Operations Research (OR) fit this scheduling problem, and enable a wide range of tourist trip planning functionalities. Using the Orienteering Problem (OP) and its extensions to model the tourist trip planning problem, allows to deal with a vast number of practical planning problems. language of the presentation: Japanese | ||
| 中島 章敬 | 1351076: M, 1回目発表 | 視覚情報メディア |
|
title: Automated Video Looping with Progressive Dynamism
abstract:Recently, cinemagraphs become a popular topic of conversation. It is an image that contains within itself a living moment. Traditionally, We need a lot of work to produce cinemagraphs. In this work, we introduce an fully automatic approach to video looping. The final representation is a “Progressive video loop”, which encodes a spectrum of looping videos ranging from a static image to a highly animated scene. The representation also supports interactive user control on local scene activities. Finally, I explain our future work. language of the presentation:Japanese 発表題目:プログレッシブダイナミズムを備えるループビデオの自動生成 発表概要:近年，画像の一部分だけ動くシネマグラフが話題となっている.シネマグラフとは動く部分を含む画像である．従来，私たちはシネマグラフを制作するために手作業などの多くの労力を必要とした． 本紹介では，シネマグラフを自動的に生成する新しい手法を紹介する．プログレッシブビデオループという静止画から動画まで変動するビデオループ群を作成する. また、インタラクティブなユーザ制御も備える．最後に，今後の展望について述べる． | ||
| 増田 優 | 1351098: M, 1回目発表 | 自然言語処理学 |
|
title: Improving Dependency Parsing of English: Focusing on Feature Granularity and Long-distant Dependency
abstract: This presentation gives the methods for better syntactic parsing of English. First, we introduce some methods for improving the performance from two previous studies. Koo et al. (2008) uses word clusters, which is coarser than words but more fine-grained than POS tags, as features. Zhou et al. (2011) uses word-to-word selectional preferences acquired from the web, resulting improved long dependency relationships. Then, we explain the methods developed by the speaker's team. The method discriminates the POS tags of subordinate conjunctions and prepositions, both of which are tagged same as 'IN' in the Penn Treebank (learning data), aiming to make the granularity of the features more proper and to process long-distant dependencies better. Finally, we propose the further possibility of improvement by showing the examples from the output of our method. language of the presentation: Japanese 発表題目: 素性粒度の調整と長距離依存関係への対策による英語依存構造解析の性能向上について 発表概要: 当発表は英語の構文解析の性能向上のための手法について提示する．はじめに，性能向上のための諸手法を2つの先行研究から紹介する．Koo et al. (2008) は，その粒度が語彙と品詞タグの間に位置する語彙クラスタを素性として使っている．Zhou et al. (2011) は大規模Webデータから獲得した語の共起の選択選好を素性にし，長距離の係り受けを含めた性能改善を報告している．続いて，発表者のチームがこれまでに取り組んだ手法について説明する．これは，学習データに用いるPenn Treebankにおいて同一の品詞タグ「IN」が与えられている従属接続詞と前置詞に独立したタグを付与し直すことで，構文解析の性能が向上したというものである．すなわち，素性の妥当な粒度への変更と長距離依存への対策の一例である．最後に，発表者らの手法による解析例を挙げながら，今後の更なる改善への展望を述べる． | ||
| 赤部 晃一 | 1351002: M, 1回目発表 | 知能コミュニケーション |
|
title: Applying generalized lattices for Forest-to-String translation
abstract: In voice translation or OCR translation, errors on the recognition part are propagated to the translation part, and whole system output will have lower accuracy compare to each individual part. In previous works, to solve above problem, they translate multiple candidates of recognizer's output and translate them, then choose the best translation from each sentences. However, this method takes much time proportionally with a number of candidates. This work uses lattices of recognizer's output to solve this problem, and uses tree-based translation system to increase English-Japanese translation accuracy. language of the presentation: Japanese 発表題目: Forest-to-String 翻訳を用いた音声翻訳における Lattice の利用 発表概要: 音声認識や光学文字認識等の結果を機械翻訳する場合、認識部の誤りが翻訳部に伝播し、 システム全体の出力精度は個々のシステムを個別に利用した場合に比べて低下する。 先行研究では上述の問題点を改善するため、認識部が出力した複数の候補文を翻訳し、 スコアが最大となる訳出を選択する方法が提案されている。しかしこの方法では、 候補文の数に比例して訳出にかかる時間が増大する。本研究は、 認識部が出力した音声認識結果の Lattice を利用することで上述の問題点を改善し、 更に Tree ベースの翻訳機を利用することで日英間翻訳の精度を向上させることを目的とする。 | ||
| 角森 唯子 | 1351070: M, 1回目発表 | 知能コミュニケーション |
|
title: A dialog system that causes utterances that aid deception detection
abstract: For a long time many studies were accomplished about deception detection, but they used only information that can be obtained directly from natural dialog. However, there are a lot of cases in which humans perform dialog in a way that makes it easier to detect deception. Therefore, I propose a dialog system to get new information performing dialog in a way that makes it easier to detect deception. In this talk, I introduce previous works and my proposal method. Lastly, I present my research plan. language of the presentation: Japanese 発表題目: 発話を誘発し嘘を判別する対話システムの研究 発表概要: 古くから嘘の自動検出について様々な研究がなされているが，対話の中で自然と起こる音韻・視覚情報について議論するほかなかった．しかし，人間が嘘を検出しようとする時は、嘘が見出しやすいように対話を展開する場合も多い．そこで，対話を誘発することで不足情報をシステム自身が収集し，その情報を用いて嘘を判別する対話システムを提案する．本発表では先行研究についての紹介と手法の提案を行い，最後に今後の方針を述べる． | ||
| 奥田 裕樹 | 1351022: M, 1回目発表 | 計算システムズ生物学 |
|
title: Evaluation of transcriptome assembly platforms for RNA-Seq method using non-model organisms
abstract: Massive sequencing of mRNA using next generation sequencers have become popular method to analyze a non model organism whose genome is not open. In oder to identify genes expressed in the sample cell, it is necessary to assemble the short reads obtained by next generation sequencers and various methods and softwares have been proposed recently. In this study, we compare the size and numbers of assembled contig sequences using three different platforms and also evaluated data size to obtain gene sequence efficiently. language of the presentation:Japanese 発表題目: 非モデル生物における遺伝子発現解析のためのde novoトランスクリプトームアセンブリの性能評価 発表概要: ゲノム配列が未決定でかつ近縁種にモデル生物が存在しない生物において，mRNA配列を網羅的に再構成するde novoトランスクリプトームアセンブリは，遺伝子発現解析や二次代謝機構の解明につながる重要な技術である．選択的スプライシングや遺伝子発現量のばらつきを持つトランスクリプトームに対しては，単にゲノム配列のアセンブリ技術を応用するのは容易ではない．近年様々なアルゴリズムや機能を用いたアセンブルをするソフトウェアが多数開発され，複雑なアセンブリが可能になる一方で，結果の評価をするのが非常に難しいという問題が生じている．今回我々は，ミドリムシEuglena gracilisのRNA-Seqにおけるde novoトランスクリプトームアセンブリにおいて，サンプルのデータ量やアセンブラの種類，パラメータを変化させたときの結果を評価し，配列の統計情報や既知遺伝子との対応付けから，遺伝子発現解析などの2次解析に最適なアセンブル結果を推定した． | ||
| 真鍋 章大 | 1351100: M, 1回目発表 | 計算システムズ生物学 |
|
title: Picture improvement of perfusion imaging and real time cine scan for cardiac MRI
abstract: Cardiac has two ways of movement, breathing and heartbeat. These movement is controlled by electrocardiogram synchronization and patient stopped breathing. And these processing are usually carried out by cardiac MRI. However, the perfusion imaging that is one of the cardiac MRI imaging methods has a long imaging time, and time of the free breathing occurs. Cardiac position gap occurs at a stage of the free breathing, and there is the problem that visibility comes to have bad. In addition, the real-time cine scan used for an inspection of cardiac function can image data collection under the free breathing fast. On the other hand, there is the problem that Signal Noise Ratio and space resolving power are confined to. In this study, it is a purpose to wrestle for picture improvement by being located for these problems using warping, and revising it. language of the presentation: Japanese 発表題目: 心臓MRIにおけるPerfusion撮像とリアルタイムシネスキャンの画質改善 発表概要: 心臓は、主に呼吸による動きと、拍動による動きの２通りの動きがある。心臓MRIでは、各々の動きを患者の息止め、心電図同期を用いて抑制し、撮像することが通常である。しかし、虚血性心疾患を評価できるPerfusion撮像は、撮像時間が長く自由呼吸の時間が発生する。自由呼吸の段階で心臓の位置ずれが生じ、視認性が悪くなってしまう課題がある。また、心機能検査に用いられるリアルタイムシネスキャンは、データ収集が速く自由呼吸下でも撮像可能である。その反面、SNR（信号対雑音比）や空間分解能が制限されるという課題がある。本研究では、これらの課題をワーピングを用いてモーションコレクションすることで、画質改善に取り組むことを目的とする。 | ||
| 木下 雄貴 | 1351035: M, 1回目発表 | 数理情報学 |
|
title: Localize electrical activity in the brain for statistical segmentation of tone sequence in Macaque monkeys abstract: Human beings can learn word segmentation from continuous tone sequences based on transitional probability, and localization electrical activity in the brain has been reported. On the other hand, It is not yet well understood whether other animals can learn such segmentation and where electrical activity in the brain is. In this study, we localize brain activity in the non-human primates by decoding the information about segments from multichannel electrocorticogram (ECoG) signals. language of the presentation: Japanese | ||
| 石原 弘二 | 1351006: M, 1回目発表 | 知能システム制御 |
|
title: Evaluation of Tracking Performance of Model Predictive Control for Modeling Error
abstract: In this study, we consider using Model Predictive Control (MPC) for exoskeleton robot control. MPC is an online trajectory optimization algorithm. MPC has been considered as a computationally intensive method and cannot be applied to real time robot control. However, due to the recent rapid improvements in computer performance, MPC is becoming a popular approach to generate optimal movement trajectories in robot control. Since joint angle movements of the exoskeleton robot can be disturbed by physical interactions with users, we need an optimal feedback policy for exoskeleton robot control to cope with the disturbances. However, it is generally difficult to analytically derive an optimal controller for a high-dimensional nonlinear system such as an exoskeleton robot, and usually iterative calculations are necessary. In this talk, as a candidate control method for exoskeleton robot, we investigate how MPC works to generate an optimal feedback policy on a simulated 2-link arm model in trajectory tracking tasks. Especially, we focus on evaluating how MPC relies on accuracy of the robot model. We show that the 2-link arm tracks a target trajectory with MPC and recovers to the trajectory if disturbances are added with various amplitude. Moreover, we show that if positions of each center of mass are different between the robot and its model and characteristics of actuators are different in terms of friction, the 2-link arm behaves well with MPC. language of the presentation: Japanese | ||
| 佐藤 諒 | 1351052: M, 1回目発表 | ロボティクス |
|
title: Development of prosthetic hand with realistic appearance and human-like motion
abstract: Most of the people who lost their hands equip prosthetic hands. The cosmetic prosthetic hand is realistic-looking but not functional. The body-powered prosthetic hand is functional but not realistic-looking because of its hook. The myoelectric prosthetic hand has realistic appearance and functionality to open and close its fingers. But it is less realistic-looking than the cosmetic prosthetic hand and cannot behave naturally because of its rigid joints. So in this research, I develop the prosthetic hand that has realistic appearance and flexible joints and can behave naturally according to the will of the user. language of the presentation: Japanese | ||
| 清水 怜 | 1351056: M, 1回目発表 | コンピューティング・アーキテクチャ |
|
title:Cooperation by Multiple Ring Accelerators
abstract: In recent years, demand of large-scale calculation such as weather simulation, fluid analysis and image processing. These applications have data parallelism and are achieved improved performance using accelerators such as GPGPU. However, the existing accelerator has the data transfer bottleneck. Therefore, we have proposed an accelerator named EMAX with reusing data mechanism and hiding the data transfer mechanism. There is a limit to the size of the kernel that can be mapped to EMAX. Cooperation of multi-EMAX is required to map the large kernel. I will present an outline of EMAX and the need for cooperation of multi-EMAX. language of the presentation: Japanese 発表題目: 複数のリング型アクセラレータ(EMAX)による連携 発表概要: 近年，気象計算，流体解析，画像処理のような大規模計算の需要が高まっている。それらのアプリケーションはデータ並列性を有しており，GPGPUのようなアクセラレータを用いて性能向上を図るケースが多い。しかしながら，既存のアクセラレータの場合，データ転送がボトルネックとなり，性能を引き出すことが難しい。そこで我々は，データを再利用する仕組みと，データ伝送を隠蔽化する機構を備えたEMAXと呼ばれるアクセラレータを提案した。ところが，EMAXはマッピングできるカーネルのサイズに限りがあるため，EMAXを複数連携し，マッピング可能なカーネルのサイズを増やす必要がある。本発表ではEMAXについて紹介するとともに，EMAXの複数連携の必要性を説く。 | ||
| 西口 絢人 | 1351082: M, 1回目発表 | ソフトウェア工学 |
|
title: Classification of defects based on the code review
abstract: To fix bugs in early stage of development is important to improve the quality and cost performance of development. Also, code review is used as one of the early bug fix method in the development field. Recently, various review methods have been proposed and some replicated experiments performed to compare the efficiency of each method. On the other hand, because of review efficiency is more affected by individual difference rather than difference of method, we need to consider individual differennce of review. However, the studies that review in consider individual difference has been rarely. In this study, We aim to identify the kind of bugs that affect individual difference by analyzing the individual results of the code review. language of the presentation: Japanese 発表題目: コードレビューに基づいた欠陥の分類 発表概要: 開発のコストパフォーマンス及び品質向上のためにバグは早期に取り除くことが重要である．また，早期のバグ除去の手段の一つとしてコードレビューは多くの開発現場で使用されている． 近年では様々なレビュー手法が提案されており，手法によるレビュー効率の影響を比較する反復実験も数多く行われている．一方で，レビューの効率は手法よりむしろ個人差に大きな影響を受けることが知られている． そのため、レビュー効率の個人差を考慮する研究が求められるが，実際にはあまり行われていない． 本研究では，個人のコードレビューの結果を比較，分析することで個人差に影響するバグの種類を特定することを目的とする． | ||
| 鈴木 貴裕 | 1351061: 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: Construction of driver models for comfortable driving routes recommendation
abstract: Features while driving on a road such as fuel consumption and driving comfort are in general different among roads and/or drivers. To recommend a personalized and driving-comfortable route to the destination taking into account road features and the driver's skill, it is desirable to predict the features while the driver drives on an arbitrary road interval. To this end, we conducted an analysis of the driving logs collected from 7 subjects. In this presentation, we report the result of the analysis. language of the presentation: Japanese 発表題目: 快適ドライブ経路推薦に向けた運転者モデルの構築 発表概要: ブレーキング，アクセル，ステアリングコントロールに起因する燃費や乗車快適度などの走行時特性は，一般に，走行する道路や運転者の技能によって異なる．運転者の目的および運転技能に 合わせて目的地までの快適な経路を推薦するためには，任意の道路区間における運転者の走行特性を予測できることが望ましい．本発表では，運転者の 走行特性を予測するための運転者モデルを構築することを目指し， 7人の被験者から収集した走行データの分析結果について述べる． | ||
| 大沢 達矢 | 1351015: M, 1回目発表 | 情報基盤システム学 |
|
title: [Paper Introduction]Peer-to-peer service provisioning in cloud computing environments
This paper aims to advance the management and delivery of services in large, heterogeneous, uncertain, and evolving cloud computing environments. The goal is important because such systems are becoming increasingly popular, yet existing service management methods do not scale well, and nor do they perform well under highly unpredictable conditions. If these problems can be solved, then Information Technology (IT) services can be made to operate in more scalable and reliable manner. language of the presentation: Japanese | ||
| MEHNAZ SERAJ | 1351122: M, 1回目発表 | 情報基盤システム学 |
|
title: A survey on modified on demand routing protocol for Mobile Ad hoc Network (MANET)
abstract: How to route packets efficiently in MANET is a very important problem. A variety of extensions/solutions in wireless routing algorithms have been proposed in the literature. A summary is given about recently proposed modified on demand routing protocols which have emphasized on different performance issues like ( mobility awareness, maximize packet delivery ratio, solving link failure, load balancing) etc. In future, I would like to analyze the routing algorithms performance when they will use various performance metrics while taking routing decision using Markov decision tool. language of the presentation: English | ||