| 荒木 貴光 | 1061201: D, 中間発表 | 池田 和司, 松本 裕治, 竹之内 高志 |
|
title: Adaptive Markov Chain Monte Carlo for Auxiliary Variable Method and Its Application to Parallel Tempering
abstract: This study solves a parameter setting problem of Auxiliary variable methods. Auxiliary variable methods such as the Parallel Tempering and the Cluster Monte Carlo method generate samples that follow a target distribution by using proposal and auxiliary distributions. In sampling from complex distributions, these algorithms are highly more efficient than the standard Markov chain Monte Carlo methods. However, their performance strongly depends on their parameters and determining the parameters is critical. Therefore, we proposed an algorithm for adapting the parameters during drawing samples and proved the convergence theorem of the adaptive algorithm. We applied our algorithm to the Parallel Tempering. That is, we developed adaptive Parallel Tempering that tunes the parameters on the fly. We confirmed the effectiveness of our algorithm through the validation of the adaptive Parallel Tempering, comparing samples from the target distribution by the adaptive Parallel Tempering and samples by the conventional Parallel Tempering. language of the presentation: Japanese | ||
| 星野 真人 | 1061204: D, 中間発表 | 池田 和司, 金谷 重彦, 銅谷 賢治 |
|
Title: Adaptive selection of learning strategy for autonomous sequence
learning in rats
Abstract: This study addresses how animals select the right strategies for motor sequence learning without any prior knowledge of correct sequence or task constraints. We developed a motor sequence learning task in which rats learn to press levers on the walls of an octagonal operant chamber in an unknown from intracranial self-stimulation reward only after the completion of the entire sequence. Two rat could learn three-step sequence among four levers with more than 80% accuracy. We observed non-monotonic changes in their lever press sequence features. Correlation analysis between the motor sequences of rats and those from reinforcement learning simulations by the Q-learning algorithm with different length of memory indicated that the state variables and learning parameters changed with the progress of learning. These results suggest that rats can flexibly choose appropriate learning methods and parameters in accordance with learning progress. Language of the presentation: Japanese | ||
| To Thi Chinh | 1051206: M, 2回目発表 | 松本 裕治, 中村 哲, 新保 仁, 小町 守 |
|
title: Analysis of patterns of relative clauses in complex sentences using syntactic information and functional words for Statistical Machine Translation
abstract: This study tries to deal with machine translation (MT) of complex sentences from English to Japanese. In order to overcome the problem of the differences in word orders between the source and target languages, we take an approach to divide complex sentences into simple clauses for MT, using syntactic information and functional words. Assuming that there is a little variation in syntactic structure of connectivities in complex sentences, we try to extract a set of high frequency patterns of relative clauses based on the syntactic information of relative clauses (named as SBAR in Berkeley Parser) in complex sentences,and use these patterns as clues to divide complex sentences into simple clauses.We have, for example, got a set of 150 SBAR patterns which cover 86% of all relative clauses. Also, as relative clauses are closely related to function words, which help to define the pattern and the relations among the components of a sentence, the study of SBAR patterns may help the translation of function word structures. language of the presentation: English | ||
| 王 昊 | 1151209: M, 1回目発表 | コンピューティング・アーキテクチャ |
|
language of the presentation: Japanese 発表題目: ***多様なアクセスパターンに適応する アクセラレータ向けメモリアクセス機構 発表概要: 我々は，画像処理や科学技術計算に向けの線形アレイ型アクセラレータLAPP（Linear Array Pipeline Processor）を提案している．LAPP は最内ループの命令列を マップ機構により演算器アレイに写像する手法を用いて，プログラムを高速実行でき る．しかし，LAPP は様々なプログラムにおける多様なメモリアクセスパターンに柔 軟に対応できないという問題ある．また，演算器アレイの規模に比例して，メモリデー タを演算器に供給するためのデータパスが増加してしまう．本稿では，多様なメモリ アクセスを実現するEMAX（Energy-aware Multi-mode Accelerator eXtension） を提案する．EMAX は，プログラムのメモリアクセスパターンを考慮した上で，分 散配置した中容量のキャッシュメモリを分割し，シフトレジスタと組み合わせること で，物理的にはシングルポートメモリでありながら，論理的には多ポートメモリを実 現する．本提案メモリアクセス機構により，メモリデータを演算器に供給するための データパスを大幅に削減するだけでなく，同一プログラムであっても，LAPP と比較 して，最内ループの命令列を削減し，演算器アレイの段数を削減できるため，更に高 性能と低電力な新型演算器アレイアクセラレータを実現できると考えられる． | ||
| 関 賀 | 1151211: M, 1回目発表 | コンピューティング・アーキテクチャ |
|
title: Defining an instructive configuration ASM for a host/accelerator framework
abstract: General purpose processors and accelerators are different processing units and have their own strengths for different kinds of programs with specific behaviors. In this research, a host/accelerator framework is proposed to best exploit the energy efficiency for a single program with various behaviors inside it. In order to cooperate the two parts, I am working on designing an instructive configuration ASM for a host/accelerator framework to instruct the accelerator to work on the hot loops inside the program. Two types of instructions have been designed with one for data transferring and the other for configuring the data graph. language of the presentation: English | ||
| 徐 浩 | 1151213: M, 1回目発表 | コンピューティング・アーキテクチャ |
|
title: Translating a 32bit processor simulator into an emulator with a self-defined ISA
abstract: Recently, new materials are being explored to work as an alternative solution for the semiconductors. However, before they can be really used for electronic devices and microprocessors, methods are required to test the usability at a relatively small scale. An 8-bit processor will be used for this purposed and in order to extend the test-ability, the 8-bit processor is going to be designed to be able to work on normal 32-bit programs. In this research, I am designing a 32-bit emulator for this purpose. Specifically, a 32-bit ARM simulator which has all functions of a 32-bit ARM processor will be transferred based on an 8-bit specialized instruction set architecture (ISA). language of the presentation: English | ||
| Budi Rahmadya | 1151206: M, 1回目発表 | ソフトウェア基礎学 |
|
title: A Method for Improving Data Delivery Efficiency in Delay Tolerant VANET
With Scheduled Routes of Cars
abstract: This paper proposed a message delivery method in a vehicular ad hoc network (VANET). It is difficult to deliver a message stably in a VANET because of the frequent link disconnections. In this paper, the authors proposed a "carry, store and forward"-based method aiming for a high delivery ratio and low message overhead. Each car periodically exchanges control information with neighboring cars. The control information includes the car's current position and scheduled driving route in its navigation system. Based on the exchanged information, each car forwards its messages to the neighboring car that will approach the closest to the destination. By simulation experiments, the proposed method achieved a better delivery ratio and lower bandwidth usage than Geo-cast routing and other epidemic routing methods. However, the long delivery time is still a problem in the method. Thus, to improve the method, we plan on using multi-hop communication to collect control information not only from the neighboring cars but also from all of clustered cars. language of the presentation: English | ||
| 藤原 雄介 | 1151204: M, 1回目発表 | インターネット工学 |
|
title: Introduction of a paper "Traffic Dispersion Graph Based Anomaly Detection" abstract: Anomaly detection of network traffic in Internet is an important issue. I will introduce a paper "Traffic Dispersion Graph Based Anomaly Detection" which proposed a new method for traffic anomaly detection. In this method, a graph, called Traffic Dispersion Graph(TDG) is generated by the network traffic. The property of TDG reflects the traffic anomalies and they can be detected by the TDG. language of the presentation: Japanese | ||
| 大木 浩武 | 1151202: M, 1回目発表 | ユビキタスコンピューティングシステム |
|
title: On the user interface for the energy-saving support system taking into account users comfort levels
abstract: In recent years, people have been interested in energy saving to reduce the electricity consumption. Most of the electricity energy consumption of households is by home appliances such as air conditioner, light, and fridge. There are many studies that try to reduce the energy consumption by home appliances, but most of them do not take into account preservation of user's comfort level when saving energy. Too much energy saving is likely to cause deterioration of user's comfort level, leading to difficulty of long term energy saving activity. In order to realize a home appliance control system with energy saving support while preserving user's comfort level, user interface which facilitates user and system to make a compromise, is essential. In this study, we propose a user interface which facilitates easy change and adjustment of energy consumption for each appliance depending on the desirable comfort level or the target energy consumption amount. language of the presentation: Japanese 発表題目: 快適性を考慮した省エネ行動支援システムにおけるユーザインタフェースの提案 発表概要: 近年，電力消費量の削減，省エネに関する人々の関心が高まっている．一般家庭における電力消費の大部分を占めているのは，エアコン，照明，冷蔵庫といった生活家電である．生活家電の電力消費を抑える研究は様々な研究がなされているが，これらの研究の多くは利用者の気分や状況が考慮されていない．ユーザにとって快適でない省エネ方法は，導入したとしてもユーザの不満が募ることにより長期間の運用が困難である．ユーザの快適性を維持しながら省エネ目標を達成するための家電制御システムを考えたとき，ユーザとシステムが歩み寄るためのインターフェースが必須となる．本研究では，ユーザとシステムの相互理解を円滑に進める方法として，快適性，省エネをより高めたいときに，各家電に使用する電力量の割合の決定や家電の設定を容易に行えるようなインタフェースを提案する． | ||
| 楊 欣 | 1151215: M, 1回目発表 | ソフトウェア設計学 |
|
title: Paper Introduction "Social Networks in Software Process Improvement"
abstract: Software Process Improvement (SPI) in small organizations is often problematic since communication and knowledge sharing is more informal. In this paper, the authors have investigated a small company SmallSoft, using the framework of social network analysis (SNA) to understand the underlying structures of communication and knowledge sharing between developers and managers. Using SNA techniques, they show how management can utilize the findings. My interest in this paper includes the study of community structure using SNA to get socio-technical aspects and its potential to improving SPI activities. I propose to study the developers' communication networks in open source projects. language of the presentation: English | ||
| 谷 宗一郎 | 1151067: M, 1回目発表 | ソフトウェア工学 |
|
title: Information Integration between Development Support Systems
abstract: The goal of my study is to make software maintenance more efficient, I propose a method to associate the similar information in each development support system automatically. Many software development projects use some individual development support system such as bug tracking system, version control system, and mailing list system. When developers note the related information in each development support system, they attach the associated tag with the information by themselves. For example, when developers change some source codes to fix bugs, they note the source codes' name in the bug report and note the bug ID in the source codes (commit message). However, many developers have not attached the associated tag. Therefore, it is difficult to track the source codes which were fixed in past times, when developer detect the similar bug. In this study, I built a system to extract the associated information from each development support system automatically. I will talk about the progress and the future work of my study. language of the presentation: Japanese 発表題目: ソフトウェア開発支援システム間の情報統合 発表概要: 本研究では，ソフトウェア開発における保守作業を効率化するために，複数の開発支援システムが各々で管理する情報を自動的に関連づけることを目的とする．ソフトウェア開発プロジェクトでは，不具合票やソースコードを一元管理するバグ管理システムや版管理システムを異なる開発支援システムとして使用していることが多い．開発者は複数のシステムに関連情報を登録する場合，登録内容に手動で関連タグを付する例えば，不具合修正のためにソースコードを変更した場合，不具合票にソースコード名，及び，ソースコード（コミット記録）には不具合番号を付する．しかしながら，開発者の多くは関連タグを付しておらず，修正内容の追跡が困難となっている．本研究では，不具合票やコミット記録などのテキストに含まれる単語を用いて，各システムから自動的に関連情報を抽出するシステムを構築した．本発表では，その進捗と今後の課題について述べる． | ||