日時: 9月24日(月)3限 (13:30-15:00)

会場: L1

司会: 小町 守
愛須亮太 1151001: M, 2回目発表 池田和司, 松本裕治, 川人光男
title: Reconstruction of NIRS Signal from fMRI Data to Validate VBDOT Algorithm.
abstract: NIRS is a portable and non-invasive functional brain imaging spectroscopy. VBDOT (Variational Bayesian Diffusion Optical Tomography) is a method of accurate three dimensional reconstruction considering with sparse assumption of brain activity on NIRS. VBDOT were proved quite effective in Phantom experiment, but the validation of human actual experiments are under investigation. In this presentation, we compared with the difference between reconstructed NIRS signal from fMRI data and actual NIRS signal for validation of VBDOT algorithm on human data.
language of the presentation:Japanese
中村俊輔 1151076: M, 2回目発表 池田和司, 松本裕治, 船谷浩之
title: Strategy Estimation of American Football by Machine Learning
abstract: Estimation of opponent team's strategy is important to win american football game, but it takes so long time and much money because the analysis for estimation is made by hand. The goal of our research is analyzing and estimating opponent team's strategy by machine learning.
language of the presentation: Japanese
山内祐輝 1151110: M, 2回目発表 中村哲, 松本裕治, 戸田智基, サクリアニサクティ, グラムニュービッグ
title:Answer Sentence Generation using Relationships between Words for Guiding Users to New Topics in Spoken Dialog Systems
abstract: We propose generating a answer sentence using the template created by hand and relationships between words. In order to generate a natural answer sentence, we use the concept dictionary and the search Web. For each method, we evaluate the naturalness.
language of the presentation:Japanese
発表題目: 対話システムにおける話題誘導のための単語間の関係性を用いた応答文生成
発表概要: 本発表では,ある単語に関する話題から別の単語に関する話題へと誘導するために,テンプレートを用いた話題誘導応答文生成法を提案する.人手で用意するテンプレートに対して,誘導対象となる単語データベースの中から埋め込み対象となる単語ペアを効率的に抽出するために,概念辞書とWeb検索という情報源を活用する.また,生成された応答文に対して主観評価により自然性を評価した.
近藤修平 1161023: D, 中間発表 松本裕治, 中村哲, 新保仁, Kevin Duh
title: Hidden Markov Tree Word Alignment Model
abstract: Though sequential word alignment models are efficient and widely used today, their assumptions are sometimes inadequate for distant language pairs. On the other hand, tree-based alignment models have their limitations such as high computational cost or using only one side of syntactic structures. We propose an efficient word alignment model that uses both side of syntactic structures, based on the Hidden Markov Tree Model, and report the experimantal results on Japanese-English word alignment.
language of the presentation: Japanese

会場: L2

司会: 中島 悠太
大庭亮 1151027: M, 2回目発表 小笠原司, 横矢直和, 加賀美聡, 西脇光一
title:Design Optimization of Robotic Hands for Life Support
abstract:In this work we propose a method to optimize robotic hand for life support. When the number of fingers, degree-of-freedom or finger link length are changed, we evaluate how effective a robotic hand is. To simulate and evaluate a robotic hand, Open-source software and well-established grasp quality metrics are used. We search for the optimal design of robotic hand based on the grasp quality metrics.
language of the presentation:Japanese
落合佑哉 1151033: M, 2回目発表 小笠原司, 横矢直和, 高松淳, 竹村憲太郎
title: Handwriting Interface using Simultaneous Localization and Mapping for Mobile Robots
abstract: Movement is one of basic functions for robots, but it is a burden for operator to concentrate continuously for teleoperation. Therefor, it is necessary to reduce the load of control by user interfaces. Recently handwritten maps have been studied and proposed as novel user interfaces, but the effectiveness of handwritten map depends on operator's skills. To solve this problem, we propose a novel handwriting user interface which employs SLAM (Simultaneous Localization and Mapping) and human tracking to adapt dynamic environments.
language of the presentation: Japanese
田中康之 1151066: M, 2回目発表 小笠原司, 横矢直和, 高松淳, 竹村憲太郎
title:Categorization of places using point clouds
abstract:We expect that intelligent robots provide suitable service in each place, so it is important for robots to categorize places. Recently, multi-layerd LIDARs have been released as commercial products, so we can get point clouds widely. Therefore, In this presentation, I will propose the method which categorizes places using point clouds, and talk about the effectiveness of proposed method. The proposed method uses SpinImage as feature which is obtained from point cloud, and Bag-of-features is employed for categorization.
language of the presentation:Japanese
金谷典武 1161020: D, 中間発表 横矢直和, 小笠原司, 佐藤智和
title: Measurement of Outdoor Environments using a Composite Sensor
abstract: A 3D modeling technique for an urban environment can be applied to several applications such as landscape simulation, navigation, and mixed reality. In this field, firstly, the target environment is measured by using several sensors (laser rangefinder, cameras, GPS, and gyro). The 3D model of the environment is then constructed based on the result of the 3D measurement. In this 3D modeling process, 3D points which exist on moving objects become an obstacle to construct accurate 3D model. In order to solve this problem, in this report, we propose a method for detection of 3D points on moving objects from 3D point cloud data using omnidirectional images. In our method, 3D points on moving objects are detected based on luminance variation obtained by projecting 3D points onto omnidirectional images.
language of the presentation: Japanese
発表題目: 複合センサを利用した実環境の計測技術に関する研究
発表概要: 現実環境の測定結果に基づいて,広域な屋外環境の三次元モデルを生成する技術は景観シミュレーションや複合現実感などの様々な分野への応用が期待されている.現実環境の測定に基づく三次元モデル生成手法では,レーザレンジファインダやカメラ,GPS,ジャイロセンサなどのセンサを用いて計測されているが,三次元測定時に移動物体が含まれた場合,これらの点は三次元モデル生成の過程において妨げとなるという問題がある.この問題を解決するために,本研究では,三次元モデル生成のための前処理として,レンジファインダから得られる形状データと撮影位置・姿勢情報が既知の複数の全方位画像から得られる輝度情報を利用して,測定データ中から移動物体上の点を除去する手法を提案する.提案手法では,レンジファインダで測定した点を複数の全方位画像上に投影した際の輝度の変化に基づき,移動物体上の測定点を抽出する.

会場: L3

司会: 橋本 健二
二村阿美 1151093: M, 2回目発表 松本健一, 関浩之, 門田暁人
title:Evaluation of program understandability based randomness of instruction
abstract:In recent years,a number of program obfuscation techniques have been proposed.However,it is difficult to evaluate the degree of obfuscation.The purpose of this study is to evaluate the program understandability by quantifying the randomness of instruction.We consider that program has a worst understandability if its instructions appear totally at random,that is,(A) all instructions has an equal frequency of appearance,and (B) all instucion sequences appears at random.We quantified (A) by using the concept of entropy and (B) by the concept of Kolmogorov complexity.Furthemore,we quantified the lower bound of information that needs to be understood by a programmer using the Kolmogorv complexity.We eveluated the validity of our proposal through case studies.
盛慎 1151135: M, 2回目発表 松本健一, 関浩之, 門田暁人
title: Module classification and Data selection for Bug prediction
abstract: In recent years, software scale is increasing while development period has been shortened. Thus, it has been asked to develop reliable software in a short period of time. In order to increase the reliability of the software, software testing is essential. It is possible to improve the efficiency of software testing by predicting where bug exists in software. Therefore, various studies of bug prediction have been so far. Prediction model, used for bug prediction, becomes better one by building using software similar to the software to be predicted. In our research, we build prediction models based on two approaches, module classification and data selection by elimination of unnecessary data. It is aim to improve the prediction accuracy thereby.
language of the presentation: Japanese
発表題目: バグ予測のためのモジュール分類とデータ選定
発表概要: 近年,ソフトウェアの規模が大きくなる一方で,開発の期間は短くなってきている. そのため短期間で高信頼なソフトウェアを開発することが要求されている. ソフトウェアの信頼性を高めるためにはソフトウェアテストが不可欠である. バグが含まれている部分を重点的にテストすることでテストの効率化を図るために,ソフトウェアのどの部分にバグが含まれているかを予測する研究が行われてきた. バグ予測を行う際には予測モデルを立てることになる. 予測対象ソフトウェアに類似するソフトウェアをモデル構築に使うことで,精度の良いモデルを構築することが出来る. 本研究では,モジュールを「分類」,あるいは予測データに含まれる不要なデータを除去し,データを「選定」することによって,予測対象に合うソフトウェアを扱ってモデルを構築する. それにより,予測精度を向上させることを目的とする. 本発表では,「分類」や「選定」を行うことによって予測精度がどのように変化するのか,その実験の結果と考察を述べる.
平山力地 1151088: M, 2回目発表 飯田元, 関浩之, 市川昊平, 吉田則裕
title: Automatic Program Segmentation using Program Slicing for Supporting Comprehension of source code
abstract: During the software development process, software maintenance is generally considered to be time consuming and costly. Typically, in order to maintain the code, software developers, first, segment the source code into several code fragments, each of these implementing a single feature of the software; and then they understand every feature implemented in each of the code fragments individually. Assuming that a single feature is achieved by interdependent statements in the source code, we intend to support this code fragmentation by separating the source code into groups of interdependent statements. For such a purpose, we use Program Slicing for analyzing dependence among statements. First, a tool was implemented to apply the proposed approach. Then, we applied it to different examples of source code methods, which included multiple features. Last, we confirmed that each of the suggested code fragments by our tool actually implemented a single feature in a method. In this presentation, we talk about the technique used to find a single feature using Program Slicing, and provide some examples of our case study.
language of the presentation: Japanese
発表題目: スライスを用いた分割に基づくプログラム理解支援手法
発表概要: ソフトウェア開発では,ソフトウェアの保守に多くの時間とコストが費やされている.開発者は保守作業において,ソースコードを単一の機能を実現するコード片に区切りながら,各コード片が実現している機能を理解する.本研究では,単一の機能はソースコード中の複数の文が変数を介して互いに協調し合って実現されるとし,協調し合うコード片の集合単位にソースコードを区切ることによって理解支援を行った.文間の計算結果の受け渡しの分析には,プログラムスライス技術を用いる.提案手法をツールとして実装し,複数の機能を含むメソッドに対してケーススタディを行い,提案手法により機能と考えられるコード片を検出できる場合があることを確認できた.本発表では,プログラムスライス技術を用いて単一の機能を検出する手法と,ケーススタディの例を紹介する.
米澤拓吾 0861022: D, 中間発表 松本健一, 関浩之, 飯田元, 門田暁人
title:Evaluation of Visual Properties in Japanese Instructional Manuals.
abstract:In designing manuals, the visual impression of the text is extremely important in motivating users to read the manual. To empirically clarify how visual properties of textinfluence the legibility and provide a quantitative guideline for designing user-friendly manuals, we manipulated four visual properties of Japanese text using Universal Design Fonts, i.e., the size of characters, the aspect ratio (width-to-height ratio) of each character, and the space between lines and characters. We experimentally evaluated how these properties affect the legibility of texts, using a conjoint method and a rating method with a rating scale of 21 steps. Severalthresholds were found that influenced the legibility of visual format.
language of the presentation:Japanese
発表概要:日本では取扱説明書の制作における書式の基準が必ずしも十全でない.例えば,日本工業規格JIS S0137『消費生活用製品の取扱説明書に関する指針』では取扱説明書で用いる文字の大きさについて「用いる活字の大きさは,3.2mm(9ポイント)以上5.6mm(16ポイント以下)が望ましい」としているが,現状では,多くの取扱説明書がこの指針の推奨値を満たしていないのは明らかである.同様に,行間,文字間隔についてはこの規格では全く触れられていない.またマニュアルの書式はメーカーのコスト的な制約のために必ずしもユーザーにとって読みやすいものとはなっていない.いわばコストと読みやすさとの間でトレードオフの関係が成立している.そこで本研究では,取扱説明書で用いられている書式(文字の大きさ,行間,文字間,文字の縦横比から構成される)にたいする読者ユーザーの選好度と書式の視覚的な読みやすさについてコンジョイント法等を用いて検証し,具体的に文字サイズ,行間,文字間,縦横比について,読みやすさと読みにくさを分ける指標を示す.