| 西川 諒 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 安本 慶一, 笠原 正治, 妙中 雄三 |
|
title:Research and study on distributed firewall based SDN data planes
abstract:Since the perimeter firewall is mainly installed between the Internet and the internal network in the organization, it cannot detect unauthorized communication in the internal network. In addition, when the internal network is configured with Openflow, Openflow cannot detect unauthorized communications. There is a method of implementing a firewall on the SDN controller, but the load is concentrated on the controller. Therefore, we aim to build a distributed firewall by extending the SDN data plane and implementing a firewall. language of the presentation:Japanese | |||
| 押田 祐冶 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 宮尾 知幸(MS), 須藤克仁 |
|
title: Anticipation for Simultaneous Machine Translation
abstract: Simultaneous translation is widely used in many scenarios including international summits, conferences, lectures and so on. Simultaneous translation is a task which begins to translate each source sentence before the source speaker finishes speaking. In this task, translation model must balance latency against quality. In previous works, the method that input a token to the model and output at each steps after waiting for fixed tokens, and the method in which reinforcement learning were proposed. Although these studies focus on the timing of starting translation, simultaneous interpreters are thought to translate while considering the outline of the story. In this study, we aim to reduce latency or increase accuracy by anticipating sentence information from halfway sentences. language of the presentation: Japanese | |||
| 廣瀨 慈恩 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子, 笠原 正治, 大下 福仁, 新谷 道広 |
|
title: Gathering in Byzantine environments with a few faults
abstract: Gathering mobile agents at the same node at the same time is a fundamental task in the field of distributed systems. This task can be more difficult to accomplish when some agents are subject to faults. In literature, a deterministic gathering algorithm for any graph with any number of agents with Byzantine faults is proposed, where agents with Byzantine faults behave arbitrarily without following their algorithms. However, the existing gathering algorithm is not practical in terms of computation time, and it is hard to imagine that many agents are subject to faults. In this study, we seek to design a deterministic gathering algorithm that relaxes only the condition of the number of agents with Byzantine faults in the same model as the literature, but that is faster than the existing gathering alogrithm. language of the presentation: Japanese 発表題目: 故障数が少ないビザンチン環境におけるモバイルエージェントの集合問題 発表概要: ネットワーク上のモバイルエージェントを同じ時刻で同じノードに集合させる動作は分散システムにおいて基本的なタスクである。 いくつかのエージェントが故障する場合、このタスクを達成することは難解になる。 既存研究として、最も悪い故障であると知られるビザンチン故障となったエージェントが任意の数存在していても任意のグラフにおいて集合タスクを達成する決定性アルゴリズムが存在する。 しかしながら、既存アルゴリズムは実用的な計算時間とは言い辛く、また、ネットワーク上のモバイルエージェントの多くが故障するとは考慮しにくい。 そこで、本研究では、既存研究と同じモデルにおいてビザンチンエージェント数の条件のみを緩和し、より高速に集合を完了する決定性アルゴリズムを提案する。 | |||
| 福田 りょう | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 須藤克仁 |
|
title: Pseudo spoken language generation for machine translation
abstract: Machine translation of "spoken language" that we routinely use in conversations is difficult because of lack of parallel data. Most of the available parallel data is “written language” used when writing a sentence, and there is very little parallel data of spoken language. To solve this problem, we can consider the fine-tuning approach that adapt the “written language” domain to the “spoken language” domain, but the difference between spoken language and written language becomes a problem. The only difference between spoken language and written language is scene where it is used, but there are significant differences in trends. For example, Japanese spoken languages tend to have simpler vocabulary than written language, have ambiguous sentence boundaries, and include fillers such as "えーと". These differences are a barrier to domain adaptation from written to spoken language. In this study, we aim to absorb these differences by translating written language into pseudo-spoken language and enable effective domain adaptive learning. language of the presentation: Japanese 発表題目: 機械翻訳のための擬似話し言葉の生成 発表概要: 我々が会話で日常的に用いる「話し言葉」の機械翻訳は難しく、その理由は学習に必要な対訳データの少なさにある。利用可能な対訳データの多くは、文章を書くときに用いられるような「書き言葉」であり、話し言葉の対訳データは非常に少ないと言える。このような対訳データの少ない領域における機械翻訳手法として、ドメイン適応が知られている。しかし、話し言葉の翻訳をドメイン適応によって学習することには多数の困難が存在する。話し言葉と書き言葉の明確な差異は「使用場面」のみであるが、その傾向には大きな違いが見られる。例えば日本語の話し言葉には、書き言葉に比べて表れる語彙が平易である、文の境界が曖昧である、「えーと」といった言い澱みが含まれる、などの傾向がある。これらの差異は、書き言葉から話し言葉へのドメイン適応の障壁となる。本研究では、書き言葉を擬似的な話し言葉に翻訳することでこの差異を吸収し、効果的なドメイン適応学習を可能にすることを目指す。 | |||
| 長谷 洋斗 | 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: Improvement of a Beam Search-Based Algorithm for Variable Order in Frontier Method
abstract: In graph problems, it is important to enumerate subgraphs that satisfy constraints such as path, spanning tree, spanning forest and so on. Frontier method is a framework to enumerate subgraphs and store them in a data structure called ZDD. Frontier method enumerates subgraphs by processing each edge one by one following a given variable order. The efficiency of frontier method depends on a given variable order. Therefore, it is important to find a variable order that accelerates frontier method. A previous study reported that a beam search-based algorithm, which is one of the heuristic algorithms, found a good variable order. In this study, we improve a beam search-based algorithm developed in the previous study. As the results of enumerating path, spanning tree and spanning forest, we show that frontier method performs almost more efficiently in the variable orders found by the proposed method than in those by the previous method for almost all instances. language of the presentation: Japanese 発表題目: フロンティア法における変数順序付けのためのビームサーチの改善 発表概要: グラフの問題において、制約を満たす部分構造を有するグラフ、例えばパス、全域木、全域森などを列挙することは重要である。フロンティア法は部分グラフを列挙し、それらをZDDと呼ばれるデータ構造に格納するための枠組みである。フロンティア法は与えられた変数順序に従って、辺を一本ずつ処理して部分グラフを列挙する。そしてフロンティア法の効率はこの変数順序に依存する。よってフロンティア法が効率よく動作するための変数順序を見つけることは重要である。これまでの研究において、ヒューリスティックスの一つであるビームサーチがよい変数順序を見つけることが知られている。本研究では、ビームサーチを改良したアルゴリズムを提案する。パス、全域木、全域森の列挙を行った結果、従来手法よりも提案手法の変数順序に従ったほうが、ほとんどの入力に対して効率よく列挙できることがわかった。 | |||
| LIU PEIMING | 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, 佐藤 嘉伸, 神原 誠之, Alexander Plopski, 藤本雄一郎 |
|
title: Finger Motor Skill Training Music Video Game Design on Smartphone
abstract: It is necessary to continue rehabilitations for a long time. But many conventional rehabilitations are monotonous and it is difficult to recognize short-term efforts. So many patients can’t continue them because they can’t maintain their motivations. We introduce a smartphone application gamification based on music play to a finger motor skill training. We believe that our system can improve the patient’s motivations and it is more effective for a rehabilitation. language of the presentation: English | |||
| 長谷川 瑛一 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 安本 慶一, 酒田 信親, 磯山 直也 |
|
title: Study on the effect of information that is always viewed to psychology and behavior
abstract: We can always view visual information from a display placed in front of our eyes by using smart glasses. However, sufficient verification has not been done on how the information that can be always viewed affects users. Therefore, in the study, we investigate what kind of contents and presentation methods of information by smart glasses brings about merit and demerit to human psychology and behavior, and aim to develop an information presentation system that can bring merit to us. language of the presentation:Japanese 発表題目: 常時閲覧情報な情報が心身に与える影響に関する研究 発表概要: スマートグラスを用いることで眼前に配置されたディスプレイから視覚情報を常時閲覧することが可能となる．しかし，常時閲覧可能な情報がユーザにどのような影響を与えるかについて充分な検証は行われていない．そこで本研究では，スマートグラスによる情報の提示内容・方法が人間の心理や行動にどのようなメリット・デメリットをもたらすのかを調査し，ユーザにメリットをもたらすことが可能なスマートグラスを用いた情報提示システムの開発を目指す． | |||
| 藤田 健太郎 | M, 1回目発表 | 大規模システム管理 | 笠原 正治, 松本 健一, 笹部 昌弘, 張 元玉 |
|
title: Mining Pool Selection Problem in the Presence of Block Withholding Attack
abstract: Mining, the process where multiple miners compete to add blocks to Proof-of-Work (PoW) blockchains, is of great importance to maintain the tamper resistance of the blockchains. In current blockchain networks, miners usually form groups, called mining pools, to improve their revenues. When multiple pools exist, a fundamental mining pool selection problem arises: how should a miner select the mining pool to join to maximize its revenue? In addition, the existence of mining pools also leads to another critical issue, i.e., block withholding (BWH) attack, where a pool sends some of its miners as spies to another pool to gain extra revenues without contributing to the mining of the latter. Although the mining pool selection has been investigated from the perspective of evolutionary game theory, the problem in the presence of BWH attack remains unexplored. Therefore, our research aims to solve this problem, i.e., the mining pool selection problem in the presence of BWH attack. language of the presentation: Japanese | |||
| 松田 明大 | M, 1回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 清川 清, 諏訪 博彦, 松田 裕貴 |
|
title:Detection of parking on the street by using drive recorder videos
abstract:Traffic accidents and Traffic jam occurred by many reasons including parking on the street. Real time sensing of parked cars is useful for estimation of dangerous area and traffic jam and so on. Previous research uses fixed-point camera to detect the parking space, however, only limited information can be obtained and costs money. Therefore, in this paper, we propose a detection method for parking on the street by using drive recorder videos. First of all, determining clear criteria for parking on the street to create the dataset. language of the presentation: Japanese | |||
| 藤石 秀仁 | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 小笠原 司, 小林 泰介 |
|
title:Continual Imitation Learning inspired by Human Learning Process
abstract: With the development of reinforcement learning (RL) and deep neural network (DNN), robots are gaining capabilities to do complex works instead of humans. RL, however, has a problem of sample inefficiently since an agent needs to visit various unknown states to find optimal actions. DNN also has a problem of catastrophic forgetting that an agent forgets the tasks learned in the past by learning new tasks incrementally. Hence, this study proposes a learning that resolves these problems. To reduce the number of samples, imitation learning, which enables an agent to imitate an expert policy from the data safety collected by the expert, is employed instead of RL. To mitigate the catastrophic forgetting, continual learning methodology, which holds the parameters in DNN important for the past tasks, is integrated to the imitation learning. With this scheme, a robot would be able to efficiently and incrementally learn various tasks. Today, a behavioral cloning from observation (BCO) was implemented as one of the imitation learning methods. This method enables an agent to imitate expert behavior with only expert trajectories, without expert actions. We confirmed the effectiveness of the method. language of the presentation:Japanese | |||
| 玉置 理沙 | M, 1回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 金谷 重彦, 藤本 まなと |
|
title:Prediction of blood glucose levels to control for non-diabetic adults
abstract:In recent years, blood glucose levels control support for diabetic patients has been actively performed, but there are few support systems for adults who are not diabetic patients. High blood sugar levels for a long time causes lifestyle-related diseases such as type 2 diabetes. In order to prevent lifestyle-related diseases, blood glucose levels control is necessary, and a system capable of predicting blood glucose levels is desirable. In this research, blood glucose levels are measured in real time using Freestyle Libre and a prediction model is constructed. language of the presentation:Japanese 発表題目:非糖尿病成人の血糖値コントロールに向けた予測モデルの作成 発表概要: 近年、糖尿病患者のための血糖値コントロール支援が盛んに行われているが、糖尿病患者ではない成人への支援システムはほとんど存在しない。長時間に及ぶ食後高血糖は2型糖尿病をはじめとする生活習慣病を引き起こす原因となり、糖尿病患者と診断される前の段階(未病)で予防することが重要である。生活習慣病を予防するためには血糖値コントロールが必要であり、リアルタイムで血糖値を把握しながら食後血糖値の予測が可能なシステムが望ましい。本研究では、Freestyleリブレを用いてリアルタイムで血糖値を測定し、予測モデルを構築する。 | |||
| 石長 篤人 | M, 1回目発表 | 情報基盤システム学 | 藤川 和利, 安本 慶一, 新井 イスマイル |
|
title: Reverse engineering of CAN for extracting each driver features
abstract:Several previous studies have shown the results of driver identification using in-vehicle network data captured in the vehicle's CAN (Controller Area Network). However, they all used signals already extracted from the DBC file, such as speed, brake pedal position, or accelerator position. Car manufacturers intentionally do not reveal the exact signal location in the DBC file. For this reason, it is difficult for third parties to reproduce this method, and DBC files for all vehicles are required when identifying multiple vehcle models. In this study, we propose CAN reverse engineering specialized in feature extraction for each driver in order to identify the driver without any information from the car manufacturer. language of the presentation: Japanese | |||