| 池上 綾乃 | M, 1回目発表 | ソフトウェア工学 | 松本 健一 | 笠原 正治 | 石尾 隆 | 畑 秀明 | Kula Raula Gaikovina |
|
title: Towards Data Analysis Assistance in Program Synthesis
abstract: Nowadays companies perform data analysis by using programs to process raw data into preprocessed data. However, reproducing these preprocessed data is a problem if the program is not longer available or the programmer has left the company.In my research, I consider how to support data analysis by using a approach to apply Program Synthesis techniques to the raw and preprocessed data.In this presentation, I will introduce two existing research techniques, investigate the benefits and problems to plan the future direction of this research. language of the presentation: Japanese 発表題目: データ分析支援に向けたプログラム合成手法の検討 発表概要: 企業では加工前後のデータは存在するが，プログラマの異動や退職，プログラムの紛失などによって過去に作成したプログラムを実行することができなくなるようなケースが存在する．そこで，本研究では加工前後のデータからプログラムを合成する手法について検討する．本発表では，既存研究のアプローチと課題点を紹介し，本研究の今後の方針について述べる. | |||||||
| 廣田 一輝 | M, 1回目発表 | ソーシャル・コンピューティング | 荒牧 英治 | 中村 哲 | 若宮 翔子 | ||
| title: *** Analysis of the content of long-term Twitter users' posts over time *** abstract: *** Social media was born in the early 2000s and is now a tool used by everyone. It has become an intimate part of people's lives and is one of the most valuable databases for understanding changes in people. In fact, however, few studies have focused on such changes in the population and analyzed them quantitatively. On the other hand, such an analysis is important for understanding future trends in social media. Therefore, this study takes up Twitter as a social media and analyzes the change in the content of Twitter users' posts over time, which is caused by the passage of more than 10 years. And accordingly, in this study, the (1) The extent to which users' statements changed, and (2) What changes were seen in the transitions in the topics mentioned, and (3) What semantic changes are seen in the words are being studied. *** | |||||||
| LIU JIA | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一 | 清川 清 | 神原 誠之 | 藤本 雄一郎 | |
|
title: Context-aware Interaction between Virtual Characters and Real Scene in Augmented Reality
abstract: How to interact more naturally with the real enviroment has always been a crucial issue in AR. In order for virtual characters to better interact with real scene, scene understanding is an indispensable part. This research is mainly divided into two parts, the first part is how to perceive and understand the scene. The proposed method is to map the results of semantic segmentation back to the scene model obtained by V-SLAM; the second part is how to design the behavior of virtual characters according to the real scene, such as decision trees or other behavioral design methods. language of the presentation: English | |||||||
| 宇野 拓磨 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清 | 小笠原 司 | 酒田 信親 | 磯山 直也 | |
|
title: Improving freedom of direction in flight experience with VR.
abstract: VR technology has made it possible to present fantasic experiences, such as flying in the sky and floating in space, and people have been enjoying it. Although many of these fantasic experiences need moveing in three dimensions, existing systems are limited in the direction and orientation of movement. In order to realize a flight experience with less restriction and more freedom of movement and posture than conventional systems, we propose a flight sense presentation system using an underwater scooter. language of the presentation: Japannese 発表題目:VRを用いた飛行体験における移動方向の自由度の向上 発表概要: VR技術により空を飛行する，宇宙空間を浮遊するといった現実空間に縛られない視覚体験の提示が可能になり，人々を楽しませてきた．こういった非現実的空間での体験は3次元的に移動するものが多いにもかかわらず、既存のシステムでは移動方向・姿勢に制限がある．こういった制限が少なく従来のシステムでは実現できなかったような移動方向・姿勢の飛行体験を実現するために，本研究では水中スクータを用いた飛行感覚提示システムを提案する． | |||||||
| 福田 晃久 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清 | 加藤 博一 | 酒田 信親 | 磯山 直也 | |
|
title: Examining Unrealistic Jumping Experiences with Visual Augmented Information
abstract: In recent years, devices such as drones and harnesses have made it possible to extend the jumping power. However, there are various limitations in providing an extended jumping experience, such as limited equipment and limited size of leap. In this study, we analyze previous researches that extended the jumping force, and aim to develop a system that provides a jumping experience that is difficult to experience in real life, such as the two-step jumping in games, by using VR and AR. language of the presentation: Japanese 発表題目: 視覚的な拡張情報による非現実な跳躍体験の検討 発表概要: 近年，ドローンやハーネスなどのデバイスなどを用いることで跳躍力を拡張することを可能にしている．しかし，拡張した跳躍体験を提供するにあたって機材設備や跳躍する大きさが限られている様々な制約が存在する．そこで本研究では，これまでの跳躍力を拡張した研究を分析し，VRやARを用いることで，ゲームで扱われる２段ジャンプなどの現実では体験しづらい跳躍体験を提供できるシステムの開発を目指す． | |||||||
| 成田 剛志 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 | 加藤 博一 | 大竹 義人 | Soufi Mazen | 上村 圭亮 |
|
title Muscle Segmentation in CT Images Using Fewer Annotated Data via Semi-Supervised Deep Neural Network
abstract Muscle segmentation is an important yet challenging task in the construction of subject-specific models in medical applications. Supervised convolutional neural network (CNN) have recently shown excellent performance, but its training requires large amount of training (labeled) data, which is time-consuming, subjective and expensive. Therefore, there is a need to obtain accurate segmentations while training CNNs with small amounts of labeled data. In this presentation, we investigate a method using semi-supervised learning based on the Mumford-Shah function, and present plans for future validation and experiments. language of the presentation Japanese | |||||||
| ZHANG BIN | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 | 向川 康博 | 大竹 義人 | Soufi Mazen | 上村 圭亮 |
|
title: Prediction of Segmentation Accuracy and Active Learning Strategy Using Bayesian U-Net-Based Uncertainty: Feasibility for Liver/Spleen Segmentation in MR Images
abstract: Bayesian U-Net has shown promising performance in image segmentation tasks. Particularly, in a previous study by our group, a Bayesian U-Net has shown possibility to segment hip and thigh muscles in CT images with a high accuracy based on epistemic (i.e. model-related) uncertainty estimation. In addition to segmenting the target structures, the estimated uncertainty demonstrated correlation with Dice coefficient. This allowed to use the uncertainty as a predictive measure for the segmentation accuracy, hence called ‘predictive uncertainty’. Furthermore, an uncertainty-based pixel selection approach demonstrated that the uncertainty predicted by Bayesian U-Net has the capability in reducing annotation cost. However, the capability of Bayesian U-Net in segmenting MR images has not been examined. In this research, we will investigate the feasibility of Bayesian U-Net both in accuracy prediction and reduction of the annotation cost in segmentation of liver/spleen in MR images. language of the presentation: English | |||||||
| 窪田 拓未 | M, 1回目発表 | ネットワークシステム学 | 岡田 実 | 林 優一 | 東野 武史 | DUONG QUANG THANG | Chen Na |
|
title: *** Analysis of transmission efficiency to multiple terminals in parallel two-wire WPT ***
abstract: *** In recent years, large scale WPT systems that supply power to multiple terminals have been attracting attention. When the transmitter size is larger than the wavelength, the voltage and current distributions are not uniform along the line. Therefore, since the received power fluctuates with the receiver's position, the It is thought that there is a positional dependence on terminals. Therefore, in this research, the transmission efficiency of multiple terminals is investigated by theoretical analysis and experiments. *** language of the presentation: *** Japanese *** 発表題目: *** 平行 2 線 WPT における複数端末への伝送特性に関する検討 *** 発表概要: *** 近年,複数端末へ給電を行う大規模 WPT システムが注目を集めている。送信器のサイズを波長よりも大きくした場合，電圧・電流分布-線路上でが一様でなくなる。そのため，受信器の位置に伴い受信電力が変動するので 端末による位置依存性があると考えられる。そこで本研究では，複数端末を用いた場合の伝送特性について理論解析および実験により検討を行う。 *** | |||||||
| 森本 康太 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一 | 中島 康彦 | 藤川 和利 | 藤本 大介 | Youngwoo Kim |
|
title:An Acceleration of Pairing Computation with RNS Representations Using an Extended Euclidean Algorithm
abstract: With the spread of IoT devices and cloud computing systems, the amount of data communicated between information devices and the number of data sharing methods have been increasing. For the background, advanced cryptography with advanced functionality such as ID-based encryption using IDs as a public key and searchable cryptography allowing data retrieval with cipher texts have been proposed. The advanced cryptography is constructed based on intensive pairing calculations and requires a fast and small-scale circuitry when used in computationally resource-intensive IoT devices. However, the algorithm of advanced cryptography is composed of intensive paring calculations requiring a large amount of computational resources. Therefore, the implementation method of advanced cryptography with high speed processing and small-scale circuit is essential for the usage in IoT devices with low computational resources. The implementation using the Residue Number System (RNS) divides large integers into smaller numbers and computes them in parallel, allowing for small and fast calculation. However, the implementation of Fermat's small theorem used for inverse calculations in pairing computation using RNS has the challenge of degrading circuit utilization. In this study, Implementation of inverse calculations using the Extended Euclidean Algorithm in pairing computation using RNS is proposed. It is confirmed that the extended Euclidean Algorithm reduce the data dependency of the inverse calculation and improve the circuit utilization, and compared the number of cycles required for the calculation. language of the presentation: Japanese | |||||||
| 中村 明星 | M, 1回目発表 | 数理情報学 | 池田 和司 | 金谷 重彦 | 吉本 潤一郎 | 福嶋 誠 | 日永田智絵 |
|
title: High-Dimensional Linear Regression for Truncated Data
abstract: We will propose a novel framework of high-dimensional sparse regression for truncated data. Truncated data is a kind of biased-sample, and constructing some good estimator for truncated regression has been widely studied since 1970s. However, most of those studies are for low-dimensional settings, and truncated regression for high-dimensional data have not been well studied. For example, previous studies have only used L1 regularization (LASSO) for extending truncated regression to high-dimensional settings, though LASSO doesn’t generally have good statistical properties. In this research, therefore, we will develop a high-dimensional truncated regression using L_{1/2} regularization, which we call truncated-L_{1/2}, and analyze its theoretical properties. L_{1/2} regularization for standard (untruncated) data has promising statistical properties, including sufficient sparsity and oracle properties, which LASSO doesn’t have. Based on these characteristics, we conjecture that our truncated-L_{1/2} would be more interpretable and credible than the existing method, truncated-LASSO. In this presentation, we are going to provide formulations of our settings and approaches, and identify what we should solve. language of the presentation: Japanese | |||||||
| 久田 祥平 | M, 1回目発表 | ソーシャル・コンピューティング | 荒牧 英治 | 渡辺 太郎 | 若宮 翔子 | ||
|
title: *** Detection of alleaging fact base defarmation ***
abstract: *** There are some cases of speech on tha internet that are legally responsible. This study focuses on the issue of speech from the perspective of legal responsibility and aims at automatic detection of problematic speech by linguistically interpreting the defamation-specific logic in it. *** language of the presentation: *** Japanese *** 発表題目: *** 事実の摘示型名誉毀損の検出 *** 発表概要: *** インターネット上には言論の問題は法的責任を負う場合がある。本研究では、言論の問題に法的責任から焦点をあて、その中の名誉毀損特有の論理を言語的に解釈することで、問題発言の自動検出を目指す。 *** | |||||||