| 平岡 拓也 | 1361011: D, 中間発表 | 中村 哲, | 松本 裕治, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title: Reinforcement Learning of Cooperative Persuasive Dialogue Policies using Framing abstract: Manually creation of a dialogue system policy is very expensive. In this report, we apply reinforcement learning for automatically learning cooperative persuasive dialogue system policies using framing, the use of emotionally charged statements common in persuasive dialogue between humans. In order to apply reinforcement learning, we describe a method to construct user simulators and reward functions specifically tailored to persuasive dialogue based on a corpus of persuasive dialogues between human interlocutors. Then, we evaluate the learned policy and the effect of framing through experiments both with a user simulator and with real users. The experimental evaluation indicates that applying reinforcement learning is effective for construction of cooperative persuasive dialogue systems which use framing. language of the presentation: Japanese | ||||||
| 赤部 晃一 | 1351002: M, 2回目発表 | 中村 哲, | 松本 裕治, | 戸田 智基, | Graham Neubig, | Sakriani Sakti |
|
title: Evaluation Framework of Machine Translation Error Analysis
abstract: Error analysis of machine translation (MT) systems are used to improve translation systems. We proposed multiple methods of MT error analysis in previous works. However, there was no method to evaluate error analysis systems of MT systems, and we could not measure accuracy of our methods. To solve the above problem, we propose a new evaluation framework of MT error analysis methods. Our framework uses corpus of machine translations and post-edit sentences which we created, and predicts error types automatically. Using our framework, we can measure accuracy of error analysis methods easily. language of the presentation: Japanese | ||||||
| 小田 悠介 | 1351023: M, 2回目発表 | 中村 哲, | 松本 裕治, | 戸田 智基, | Graham Neubig, | Sakriani Sakti |
|
title: Optimizing Segmentation Strategies for Simultaneous Speech Translation
abstract: Simultaneous speech translation system is one of application of machine translation that translates speech from the source language into the target language in realtime. In this setting, conventional translation system cannot start to translate the speech at the optimal timing, because speakers often continue their speech for a long time with ambiguous sentence boundaries. So the translation system require a segmentation strategy, which estimates the optimal sentence boundaries of the speech for the input of the translation system. We propose new algorithm to learn segmentation strategies for simultaneous speech translation. In contrast to previously proposed heuristic methods, our method finds a segmentation that directly maximizes the performance of the machine translation system. We propose a method based on greedy search and dynamic programming that searches for the optimal segmentation strategy. An experimental evaluation finds that our algorithm is able to segment the input two to three times more frequently than conventional methods in terms of number of words, while maintaining the same score of automatic evaluation. language of the presentation: Japanese | ||||||
| 波多腰 優斗 | 1351088: M, 2回目発表 | 中村 哲, | 松本 裕治, | 戸田 智基, | Graham Neubig, | Sakriani Sakti |
|
title:
Syntactic Preprocessing for Syntax-based Machine Translation
abstract: Several preprocessing techniques using syntactic information and linguistically motivated rules have been proposed to improve the quality of phrase-based machine translation (PBMT) output. On the other hand, there has been little work on similar techniques in the context of other translation formalisms such as syntax-based SMT. In this research, we examine whether the sort of rule-based syntactic preprocessing approaches that have proved beneficial for PBMT can contribute to syntax-based SMT. Specifically, we tailor a highly successful preprocessing method for English-Japanese PBMT to syntax-based SMT, and find that while the gains achievable are smaller than those for PBMT, significant improvements in accuracy can be realized. language of the presentation: Japanese | ||||||
| 小野木 祐太 | 1351027: M, 2回目発表 | 山口 英, | 笠原 正治, | 安本 慶一, | 池田 和司, | 門林 雄基, | 奥田 剛 |
|
title: Study on a Benchmarking Method of the Fault-Tolerance against Hadoop
Implementations.
abstract: The back end system of a huge web service such as Facebook uses Hadoop distribution system. Hadoop is one of implementations of MapReduce. MapReduce is a parallel programing model that has advantage of simple commodity and scalability by scale out, to process huge information data called "big data". Hadoop has been improved in various aspects data processing speed and fault-tolerance so far, but fault-tolerance have not been evaluated enough in various cases. This presentation introduces traditional benchmarking methods of fault-tolerance on the implementation of Hadoop, then I propose remediation and describe road map of my research. language of the presentation: Japanese | |||||||
| 沖 修平 | 1351021: M, 2回目発表 | 岡田 実, | 安本 慶一, | 東野 武史 | |||
|
title:Improving Detection Accuracy using Subspace Method in LCX-MIMO based Positioning System of Radio Terminals
abstract:Recently, various services using the position information in indoor environment are proposed. Accordingly, the demand for the position detection method which has high detection accuracy also indoors is increasing. The existing position detection methods have some problems in using indoors, but we can solve this by using Leaky CoaXial Cable (LCX). Therefore, we propose the LCX-MIMO system and we simulate the position detection by calculating the time difference of arrival (TDOA) using experimental data. In addition, we improve an detection accuracy of this system by using subspace method. language of the presentation:Japanese 発表題目:LCX-MIMO無線端末位置検出システムにおける部分空間法の適用による検出精度改善 発表概要:近年，屋内環境での位置情報を利用した様々なサービスが提案され，それに伴い屋内でも高い検出精度をほこる位置検出方法に対する需要が高まっている．本研究では漏洩同軸ケーブル（LCX:Leaky Coaxial Cable）を用いることで既存の位置検出手法が抱える問題を解決し，これを用いたLCX-MIMOシステムを提案する．このシステムに対して実測データを元にLCX内を伝搬する信号の到来時間差を算出し，シミュレーションによって無線端末の位置検出を行う．また，到来時間差算出の際に部分空間法を適用することによって，さらなる検出精度の改善を図る． | |||||||
| 金子 裕哉 | 1351032: M, 2回目発表 | 岡田 実, | 安本 慶一, | 東野 武史 | |||
|
title: Interference Suppression Schemes for Radio over Fiber Simultaneously Transmitted with 10 Gbps On-Off Keying Signal
abstract: Radio over Fiber (RoF) and 10 Gbps optical On-Off Keying (OOK) signal are simultaneously transmitted by using the stochastic process of OOK as a carrier for the radio frequency (RF) signal. The theoretical analysis of the power spectrum and error vector magnitude (EVM) agree with experimental results. The experimental and theoretical results show that the OOK modulation interferes with the RF signal. Interference suppression schemes are proposed and the improvement of EVM and dynamic range is discussed from theoretical analysis and simulation. language of the presentation: Japanese 発表題目: 光OOK重畳ファイバ無線における干渉抑圧法の提案 発表概要: 光OOK(On-Off Keying)信号を光源とした外部変調により，無線信号を強度変調・直接検波方式で送信する簡易な構成のRoF(Radio over Fiber)システムが提案されている．本発表ではまず提案システムで伝送する無線信号のパワースペクトルと無線信号品質の指標の一つであるエラーベクトル振幅(EVM)を理論的に解析し，実験結果と一致することを示す．実験結果と理論解析よりOOK信号が無線信号に干渉することが明らかになったため，その干渉抑圧法を提案し，EVMとシステムのダイナミックレンジへの改善効果を理論とシミュレーションによって示す． | |||||||
| 三好 崇之 | 1351106: M, 2回目発表 | 岡田 実, | 安本 慶一, | 東野 武史 | |||
|
title: A Study of Wireless Power Transfer using Parallel Line Feeder
abstract: Wireless power transfer technologies are expected to be used in many applications, for example, power feeding to portable devices, and electric vehicles.Therefore,lots of key technologies and applications are expected to be developed and proposed.This study proposes and considers a configuration of magnetic coupling in wireless power transform using parallel line power feeder. language of the presentation: Japanese | |||||||