| 山田 暉 | M, 2回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
| LIU SHANSHAN | M, 2回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
| CHOI JUNG WON | M, 2回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: The practicality of a large set of unlabeled data in text classification through multi-task learning concerning a miniscule set of labeled data
abstract: Understanding the polarity and aspect of the review text is mandatory for analyzing the consumer's opinion regarding a product. Amazon dataset and SemEval dataset are well used in sentiment analysis tasks. Amazon has a large set of review data but they are difficult to analyze as most of the data set is unlabeled. On the other hand, SemEval's review data lacks practicality because despite the convenience in analysis that it's labeled data brings, it is too small to train Deep Learning models. If we could supplement these two different sets of data with each other, it would be possible to train models based on a small set of labeling data as well as analyze a large set of unlabeled data using these trained models. We tested the above hypothesis by studying the degree of performance improvements in utilizing Amazon's large set of unlabeled review data while classifying text through SemEval's small set of review data in a similar domain, which includes sentences that have been labeled with polarity and aspect categories. We applied multi-task learning for the model training of text classification. We utilized LSTM (Long Short Term Memory) and CNN (Convolutional Neural Network) models as the most basic Neural Network model and compared their results. language of presentation: Japanese | |||
| 仲田 勇太 | M, 2回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
| 馬場 瑛義 | M, 1回目発表 | 大規模システム管理 | 笠原 正治, 井上 美智子, 笹部 昌弘, 川原 純（客員） |
|
title: On enumerating subgraphs in various graph classes with ZDD
abstract: We often regard current social problems as graph optimization problems. It is typical to enumerate substructures that satisfy certain conditions in graph problems. However, such a task is often harder due to the limitation of computational resources as the graph becomes larger. In this research, we consider a data structure called ZDD, which represents data in a compressed way, for enumerating connected components of a graph. If the instance of graphs that we try to enumerate is complex, it is hard to enumerate them with conventional ZDD. In this study, we introduce colored ZDD that is extended from conventional ZDD, and design an algorithm with which we enumerate Claw-free graphs using Colored ZDD. language of the presentation:Japanese 発表題目: ZDDを用いた様々なグラフクラスに属する分グラフの列挙に向けて 発表概要: 現在の社会的な諸問題はしばしばグラフの問題とみなすことができる。グラフに関する諸問題である、条件を満たす部分構造を列挙する問題はグラフ理論において典型的である。しかしグラフのサイズが大規模になるにつれて全てを列挙するということは計算資源の都合上困難になることが多い。本研究で はグラフの連結成分を列挙するためにZDDと呼ばれるデータを圧縮した表現方法を用いる。列挙するグラフの種類が複雑になった際、従来のZDDでは列挙することが困難となる。本研究ではZDDを拡張した色付きZDDを導入して、それを用いてClaw-free graphの列挙を行う。 | |||
| 深澤 佑樹 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元, 松本 健一, 市川 昊平, 高橋 慧智 |
|
title: Proposal of method name recommendation method based on method similarity and calling relation
abstract: Understanding source code has become an essential activity throughout the software development life cycle. In order to improve understanding of the source code, it is important to give the method name an appropriate name. However, developers often give inappropriate method names. Along with this, various methods for recommending method names have been proposed. In the existing method, the method name recommendation using the method name of the call destination is performed, but the accuracy is low. The proposed method proposes a method name recommendation method based on the method similarity and the call relation, aiming at method name recommendation with high accuracy. language of the presentation:Japanese | |||
| AO YANJIAO | 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, 神原 誠之, Alexander Plopski, 藤本 雄一郎 |
|
title: Live 360 Panorama Based Mixed Reality Collaboration System in Large-scale Environment abstract: Nowadays interactive media techniques have been considered a breakthrough for realizing tele-communication and remote collaboration. Such systems usually compose of three steps: first, the Augment Reality (AR) user captures the scene of its surroundings as a panorama video with a 360-degree camera and live it to a remote user. Second, the remote user visualizes the stream on a display; annotates on the virtual environment and then shares them back to the AR user. And third, the AR equipment receives the feedback, calculates the spatial locations of these annotations and display them on the optical see-through AR-HMD for the AR user’s reference. Different from most of the existing solutions that can only deal with a tinny-scale environment where the AR users are often assumed as static, we focus on collaborations within room-scale environment in which the users can walk around as they want. language of the presentation: English | |||
| 鶴尾 明大 | M, 1回目発表 | 数理情報学 | 池田 和司, 杉本 謙二, 吉本 潤一郎, 久保 孝富, 福嶋 誠 |
|
title:Rider’s action modeling during the horse’s jumping
abstract:A mathematical model is important for understanding a riding horse technique. The riding horse technique is an interaction between horses and riders. Regarding the interaction between horses and riders, there are spring-damper-mass systems. These models have represented the difference of the rider’s technique and have clarified biomechanical requirements of the rider well. However, the jumping of the horse and rider, which is not cyclic, remains unclear. Therefore, we will model the horse's jumping and will clarify biomechanical requirements of the rider at the jumping by using spring-damper-mass systems that are highly elaborate. First of all, we will apply these models to public video. language of the presentation:Japanese | |||
| 馬場 柾也 | M, 1回目発表 | 大規模システム管理 | 笠原 正治, 藤川 和利, 笹部 昌弘 |
|
title: A Review of Online Service Chaining Problem in NFV Networks abstract: Current network service provision relies on dedicated hardware to maintain service quality and stability. In order to reduce Capital Expenditure (CAPEX) and Operating Expense (OPEX), Network Function Virtualization (NFV) technology can improve flexibility of service provision where Virtual Network Functions (VNFs) can be chained together to support network services. One of the main challenges for the NFV deployment is resource allocation, which maps VNFs and their connections to physical nodes and links, respectively. Existing work proposes a heuristic scheme and an exact mathematical formulation to solve service chaining problem in an online manner. In this presentation, we show the existing work on the NFV service chaining problem and our future work. language of the presentation: Japanese | |||
| 荒井 公寛 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 安本 慶一, 藤川 和利, 藤本 大介 |
|
title: Fundamental Study of Secret-Key Agreement Scheme Using Environmental Electromagnetic Field Distributed Around Users
abstract: The confidentiality of information communication is ensured by cryptographic algorithms based on computational security. However, there is a risk that the cryptographic algorithm based on the computational security is compromised if the processing capability of the computer improves. Meanwhile, research on cryptographic methods using one-time pads based on information-theoretic security has been conducted, and such methods can guarantee communication security regardless of the processing capability of the computer. However, even if the encryption method is secured in terms of information theory, there are problems such as the high-cost for sharing the secret key. Therefore, in this study, we discuss a secured secret-key agreement scheme at a lower cost than the conventional methods under the condition that users share a physical characteristic. language of the presentation: Japanese | |||