| ZHANG JINGYI | 1561037: D, 中間発表 | 知能コミュニケーション | 中村 哲,松本 裕治,Graham Neubig |
|
title: Neural Network Models for Forest-to-String Machine Translation
abstract: Neural network models have become increasingly popular in machine translation recently. Our work focuses on using neural network models to improve translation quality of forest-to-string translation, which translates the n-best parse trees of a source sentence into a target sentence. We introduce three neural network statistical models into forest-to-string translation to capture different useful information for the translation process, which contain word reordering, word translating and translation rule selection. Experiments on different language pairs showed the proposed models outperformed previous related models and the combination of the proposed models achieved the best translation quality. language of the presentation: English | |||
| 石川 葉子 | 1551008: M, 2回目発表 | 知能コミュニケーション | 中村 哲,松本 裕治,吉野 幸一郎,Sakriani Sakti,Graham Neubig |
|
title: Response Selection of Emotional Expressions for Persuasive Dialog Systems
abstract: Emotional expressions is an efficient way to convey one’s thoughts each other. Especially, in persuasive situations, emotional expressions of persuader may give a strong main effect on recipients attitude. In this research, we aim to realize emotionally rich persuasive dialog system. Therefore, it is important to consider not only user’s emotional states but also system’s emotional expressions. In this study, we first collect dialog data including emotional expressions. Second, we analyze the dialog data and define system’s emotional states and dialog acts to implement into the dialog system. Finally, we construct an example-based persuasive dialog system. language of the presentation: Japanese | |||
| 笹野 仁 | 1551047: M, 2回目発表 | 知能コミュニケーション | 中村 哲,松本 裕治,吉野 幸一郎,Graham Neubig |
|
title: The Simulation of Multimodal lexical acquisition
Recently there are many researches about categorization function in brain using neuroimaging.I introduce the previous research that works on categorization using multimodal robot, and propose the method based on the simulation for expanding previous work to the dialogue system. In order to achieve the natural communication between a dialogue system and human, the similar function is required to the dialogue system as well. language of the presentation: Japanese | |||
| 平岡 類 | 1551086: M, 2回目発表 | 知能コミュニケーション | 中村 哲,松本 裕治,田中 宏季,Sakriani Sakti |
|
title: Unknown word Detection in Non-native Reading Based on Eye gaze features
abstract: Reading in Non-native language is often problematic and time costing , particularly ,when the text contains technical terms . The focus of this research is estimating user's ability and detecting problematic words measuring human gaze behaivor while reading texts. Additionally, utilizing the function of detecting problematic words, we are going to implement a function that displays simplified or translated words in response to input words. language of the presentation: Japanese | |||
| PADERNA RYAN RANARIO | 1561031: D, 中間発表 | ネットワークシステム学 | 岡田 実,杉本 謙二,東野 武史,侯 亜飛,Duong Quang Thang |
|
title: Low-Complexity Compressed Sensing-Based Channel Estimation with Virtual Oversampling for Digital Terrestrial Television Broadcasting
abstract: We proposes a low-complexity Compressed Sensing (CS)-based time domain channel estimation with virtual oversampling for OFDM based DTTB system. CS technique is first applied to sparse channel estimation in Integrated Services Digital Broadcasting-Terrestrial (ISDB-T) system. Unfortunately, the CS-based time-domain channel estimation suffers from a performance degradation under the existence of fractional delay. Fractional delay is a common issue in time domain based channel estimation due to sampling imperfection of channel delay time. Therefore, we propose virtual oversampling to deal with the fractional delay without increasing the signal bandwidth. Furthermore, exploiting the fact that the rate of change of path delay is slower than the path gain, a Modified Orthogonal Matching Pursuit (MOMP) is also proposed to reduce computational complexity. Numerical analyses of bit error rate (BER) and computational complexity verifies effectiveness of the proposed scheme. language of the presentation: English | |||
| 福見 渉 | 1551088: M, 2回目発表 | 知能システム制御 | 杉本 謙二,岡田 実,松原 崇充,南 裕樹 |
|
title: Reinforcement learning based design of quantizers
abstract: This study focuses on a model-free design of quantizers for discrete-valued input control. The key technique is a reinforcement learning algorithm. We first propose a quantizer including reinforcement learning mechanism. Then, its effectiveness is verified by a numerical simulation of feedback control system. Finally, we consider an extension of the proposed approach to dynamic quantization. language of the presentation: Japanese | |||
| 伊藤 涼介 | 1551013: M, 2回目発表 | 知能システム制御 | 杉本 謙二,安本 慶一,松原 崇充,南 裕樹 |
|
title: Mobile robot navigation by illuminance field control
abstract: This study focuses on the navigation problem of mobile robots, and proposes a navigation method based on illuminance field control. The proposed method is composed of the projection of binary images onto the field through the projector and the determination of the control input by using the information on projected images. We first formulate the design problem of two controllers: one controller is embedded in the projector and the other is embedded in the robot. Then as a solution to the problem, we derive two controllers such that the robot reaches to a given target point. Finally, we evaluate the effectiveness of the proposed method by position control experiments. language of the presentation: Japanese | |||
| 能登 健太朗 | 1551075: M, 2回目発表 | 知能システム制御 | 杉本 謙二,安本 慶一,松原 崇充,南 裕樹 |
|
title: An Approach to Irregular Sampling State Observer Design abstract: We study a state observer design when the output signal is not available regularly from the plant whose state is to be estimated. When an invalid signal is received, the observer reuses previous signal. This means a sampling period extends at this moment. However, it is unpredictable when such reuse occur, as a result, the sampling period becomes irregular, which may induce instability in the error dynamics of state estimation. To guarantee stability, we propose to design the observer gain by means of a common solution to simultaneous linear matrix inequalities corresponding to all sampling periods foreseen. language of the presentation: Japanese
| |||