| 伊藤 俊一郎 | 1551012: M, 1回目発表 | インターネット工学 |
|
title: Investigation and Consideration of Targeted Attack Detection Techniques
abstract: Targeted attacks constitute a critical threat in modern Internet security, and the detection of targeted attacks is a primary concern. Within the existing research, there is a analytical method which splits the attack by the kill chain model into seven phases. This prevents targeted attacks on the basis of the information obtained from attacks measured at each phase. In addition, there is research for identifying targeted malicious emails. It is identified by extracting features from the header and text of the email and target environments. In this presentation, I investigate and consider these existing research methods and analyzing their problems to propose a detection technique that suppressing the damage of targeted attacks. language of the presentation: Japanese | ||
| 田中 大樹 | 1551060: M, 1回目発表 | ソフトウェア設計学 |
|
title: [paper introduction] Recommendation Technique for Targets of Refactoring Using Machine Learning
abstract: Refactoring is a technique for restructuring an existing body of code, altering its internal structure without changing its external behavior. It is a very important technique for many situations in software development. Study on refactoring have been actively conducted. Several studies suggested recommendation technique for targets of refactoring to developer. Recommending targets of refactoring is useful for reducing developer's labor and time for detecting it. This presentation introduces effective recommendation technique for targets of refactoring using machine learning. language of the presentation: Japanese | ||
| KARIM MD. REJAUL | 1551125: M, 1回目発表 | ソフトウェア設計学 |
|
title: What’s in a Bug Report?
abstract: Bug reports are the primary means by which users of a system are able to communicate a problem to the developers, and their contents are important to developers fixing the bug. This paper aims to investigate how users report bugs in systems: what information is provided, how frequently, and the consequences of this. The study examined the quality and quantity of information provided in 1600 bug reports drawn from four open-source projects recorded what information users actually provide, how and when users provide the information, and how this affects the outcome of the bug. The paper demonstrates a clear mismatch between the information that developers would wish to appear in a bug report, and the information that actually appears. language of the presentation: English | ||
| 藤野 啓輔 | 1451090: M, 2回目発表 | 松本 健一，井上 美智子，伊原 彰紀 |
|
title: Selection of the versions of OSS library based on trend
abstract: Open source software (OSS) library is a suite of data and programming code that is used to develop software programs and applications. OSS library generally consists of pre-written code, classes, procedures, scripts, configuration data and more. OSS library is absolutely necessary to develop proprietary software and OSS in shorter period. To use OSS library, we need to select a reliable library from many versions that OSS library project has released. While a newer release version may have additional functions, the version is likely to be detected new issue. To identify more reliable software, I conduct an empirical study of library version trend, and mine the changes popular library version in the trend. language of the presentation: Japanese | ||
| 岩元 文 | 1551016: M, 1回目発表 | 自然言語処理学 |
|
title: Developing Japanese Morphological Analyzer robust to Unknown Words
abstract: Morphological analysis is important preprocessing for sentences which do not have explicitly marked word boundary such as Japanese. By improving morphological analyzer, we can expect accuracy of other NLP tasks e.g. machine translation to be improved. In this work, we investigate perceptoron model using local feature and deeplearning model to improve morphological analyzer. language of the presentation: Japanese | ||
| NURUL FITHRIA LUBIS | 1551128: M, 1回目発表 | 知能コミュニケーション |
|
title: Study of emotion in speech for spoken dialogue system
abstract: Advancements in spoken language technologies have allowed users to interact with computers in an increasingly natural manner. However, most conversational agents or dialogue systems are yet to consider emotional awareness in interaction. To consider emotion in these situations, social-affective knowledge in conversational agents is essential. We propose a corpus-based emotion-guidance dialogue system that is aware of user's emotional state and has the ability to guide it to another state. The components of the dialogue system are described and discussed. language of the presentation: English | ||
| 橋岡 佳輝 | 1551079: M, 1回目発表 | 視覚情報メディア |
|
title: [paper introduction] FlowNet: Learning Optical Flow with Convolutional Networks
abstract: Convolutional neural networks (CNNs) have recently been very successful in a variety of computer vision tasks, especially on those linked to recognition. Optical flow esti- mation has not been among the tasks where CNNs were suc- cessful. In this paper we construct appropriate CNNs which are capable of solving the optical flow estimation problem as a supervised learning task. We propose and compare two architectures: a generic architecture and another one including a layer that correlates feature vectors at different image locations. Since existing ground truth datasets are not sufficiently large to train a CNN, we generate a synthetic Flying Chairs dataset. We show that networks trained on this unrealistic data still generalize very well to existing datasets such as Sintel and KITTI, achieving competitive accuracy at frame rates of 5 to 10 fps. language of the presentation: Japanese | ||
| 岩口 優也 | 1551015: M, 1回目発表 | 光メディアインタフェース |
|
title:capturing clear image in bad weather utilizing ToF camera
abstract: Recently, security cameras are increasing, however captured image becomes unclear in bad weather. Because particles like fog and raindrop scatter rays. To capture clear image, we utilize Time of Flight camera. It can get distance by measuring the time of reflected rays from an object. Applying this theory, we can capture rays travelling given distance. The experiment by simulation will be introduced. language of the presentation: Japanese | ||
| 則兼 卓人 | 1551076: M, 1回目発表 | 環境知能学 |
|
title: [paper introduction] Random k-Labelsets for Multi-Label Classification
abstract: A simple yet effective multi-label learning method, called label powerset (LP), considers each distinct combination of labels that exist in the training set as a different class value of a single-label classification task. The computational efficiency and predictive performance of LP is challenged by application domains with large number of labels and training examples. In these cases the number of classes may become very large and at the same time many classes are associated with very few training examples. To deal with these problems, this paper proposes breaking the initial set of labels into a number of small random subsets, called labelsets and employing LP to train a corresponding classifier. The labelsets can be either disjoint or overlapping depending on which of two strategies is used to construct them. The proposed method is called RAkEL (RAndom k labELsets), where k is a parameter that specifies the size of the subsets. Empirical evidence indicate that RAkEL manages to improve substantially over LP, especially in domains with large number of labels and exhibits competitive performance against other high-performing multi-label learning methods. language of the presentation: Japanese | ||
| 森田 侑介 | 1551108: M, 1回目発表 | 計算システムズ生物学 |
|
title: Presumption of ovipositional stimulants and deterrents of umbelliferous medical plants for Papilio machaon abstract: Many insects feed only on specific plants. This feature is called host plant specificity. The host plants of P. machaon (common yellow swallowtail) are Umbelliferae plants. In this research, we observed six umbelliferous medical plants we cultivated. P. machaon laid eggs on some of them but not all. We gathered volatile compounds emitted from leaves of each plant and measured them with Gas Chromatography Mass Spectrometry. The result is there were chemical compounds with significantly different amount between the two groups. We plan to investigate ovipositional stimulants and deterrents and try to investigate their interaction using statistical analysis. language of the presentation: Japanese 発表題目：キアゲハに対するセリ科薬用植物の産卵刺激物質および阻害物質の推定 発表概要：多くの昆虫は特定の植物だけを食物として利用する。これを寄主植物特異性という。キアゲハはセリ科の植物を寄主とする。 今回、6種類のセリ科薬用植物の栽培を行っている中で、キアゲハが産卵を行うグループと行わないグループがあることが観察された。 各植物の葉から放 出される揮発性物質を採取し、ガスクロマトグラフ質量分析計で測定したところ、 両グループ間で含有量が顕著に異なる化合物の存在が明らかとなった。 今後、 統計解析を用いて産卵刺激物質や阻害物質の探索、およびそれらの相互作用を推定する。 | ||
| 福井 友季也 | 1551087: M, 1回目発表 | ロボティクス |
|
title: Operation Support of Unmanned Aerial Vehicle based on the Instruction of Camera Moves
abstract: Recently, unmanned aerial vehicle (UAV) is attracting attention in the fields of inspection and security. However, it is very difficult for operators to operate a UAV by considering the collision of obstacle and the state of UAV simultaneously in a complex environment. In this research, an operation support system will be proposed to operate the UAV only by giving the instructions of camera moves. 3D SLAM will be used for automatic collision avoidance and path planning based on the directions of camera. language of the presentation: Japanese | ||
| 福見 渉 | 1551088: M, 1回目発表 | 知能システム制御 |
|
title: Reinforcement Learning Quantizer for Discrete-Valued Input Control
abstract: In this presentation, we focus on the model-free design of quantizer for discrete-valued input control. The key technique of this study is reinforcement learning. We first propose a quantizer with a reinforcement learning mechanism, which is called reinforcement learning quantizer. Then, its effectiveness is verified by a numerical simulation with a nonlinear system. language of the presentation: Japanese | ||
| 古庄 泰隆 | 1551096: M, 1回目発表 | 数理情報学 |
|
title: Information Theoretical Analysis of Deep Neural Network Representations
abstract: Process of training deep neural network is divided into two parts: (1)pre-training and (2)fine-tuning. In some conditions, pre-training helps performance of deep neural network. However, why, how and when pre-training help deep neural network are little clarified. To tackle this problem, we considered a representation of each layer as a code, and evaluated entropy and mutual information of these codes. We explained why pre-training help deep neural network by comparing these information theoritical variables of network with pre-training and ones without pre-training. And, we defined properties of good representations for a given task: (1)code length (variation of code) is small and (2)containing large information for a given task. These properties are trade-off. Therefore, there might be minimum length of code for containing arbitrary amount of information for a task. We exmamined how codes after pre-training and after fine-tuning approach this limit of code, and explained how pre-training helps deep neural network and relationship pre-training and fine-tuning. At last, when pre-training helps deep neural network? Based on results of this study, we hypothesized the answer to this question. This validation is future work. language of the presentation: Japanese | ||