| ARAYA KIBROM DESTA | D, 中間発表 | 情報基盤システム学 | 藤川 和利, | 門林 雄基, | 林 優一, | 新井 イスマイル | |
| MOHD RUZEINY BIN KAMARUZZAMAN | D, 中間発表 | サイバーレジリエンス構成学 | 門林 雄基, | 笠原 正治, | 林 優一, | 妙中 雄三, | FALL DOUDOU |
|
title: *** Civil Aviation’s Cyber-Physical Systems Resilience: Enhancing Operational Resilience Through Impact Analysis of ADS-B Based Cyber Attacks ***
abstract: *** Automatic Dependent Surveillance-Broadcast (ADSB)’s openness and unencrypted nature makes it vulnerable to cyberattacks such as spoofing which may cause flight arrival or departure delays due to confusions experienced by Air Traffic Control (ATC). Further cascading delays affecting number of aircraft would occur within the ‘Arrival-Ground Movement-Departure’ (AGMOD) dynamics of an airport and synchronous disruptions would also be faced by the airport operations side. In understanding the impacts and forming appropriate mitigation measures, this research shall model the Air Traffic Controller (ATC)’s responses during ADS-B based attacks and the repercussions on Air Traffic Management (ATM) and airport operations. A thorough impact analysis will be carried out using specially developed simulators of discrete events and suitable impact assessment tools to identify possible impacts at interdependent common critical nodes of the cyber-physical systems comprising the ATC and the airport. Among the potential contributions of this research would be a model to predict impact level to the airport operations based on the advance predictive modelling of the cascading effects and an optimization scheme for the AGMOD Dynamics by alleviating attack impacts through a network of aligned queues. *** language of the presentation: *** English *** 発表題目: *** この部分を発表題目に *** 発表概要: *** この部分を発表概要に *** | |||||||
| MA YUE | M, 1回目発表 | 情報基盤システム学 | 藤川 和利, | 笠原 正治, | 柴田 直樹 | ||
|
title: Joint traffic signal control for vehicles and pedestrians at intersection abstract: Traffic signal control is an important and challenging real-world problem, which aims to minimize the travel time of road users by coordinating their movements at the road intersections. And there is almost no research on traffic signal control for pedestrian. Even if autonomous vehicles will dominate in the future, the traffic lights for pedestrians will still be needed. In my research, I would like to propose a method of joint traffic signal control for both vehicles and pedestrians at intersections using Back-pressure algorithm. It aims to improve pedestrian passing efficiency, decrease the travel time of them, and ensure the road safety of both vehicles and pedestrians. language of the presentation: English | |||||||
| 有薗 舜 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子, | 池田 和司, | 大下 福仁, | 新谷 道広 | |
|
title:
Continual Learning Based LSI Test for Lot Production Management
abstract: Large-scale integration (LSI) circuits are indispensable components of infrastructures to support the smooth functioning of our daily life, and thus their dependability is becoming enormously crucial with the spread of LSIs. To guarantee the reliability of the LSIs, LSI testing plays an important role. Recently, machine-learning-based LSI testing has been attracting attention to reduce test costs and improve test quality. However, the most conventional methods suffer from catastrophic forgetting when applied to real-world LSI test applications, where a dataset arrives sequentially under lot production management. To overcome this issue, we propose a novel sequential-learning-based LSI testing that exploits continual learning. language of the presentation: Japanese 発表題目: ロット生産管理のための継続学習を利用した集積回路テスト手法に関する研究 発表概要: 集積回路は，社会インフラに必要不可欠であり，その信頼性は非常に重要である．集積回路の信頼性を保証するために，集積回路の品質テストは重要な役割を果たす．近年，機械学習を集積回路のテストに応用することで，テストにかかるコストを削減する試みが多くなされている．しかし，既存手法の多くはバッチ処理を前提としており，データセットがストリーミングに到着するような製造工程の実情を考慮していない．またストリーミングにデータセットが到着するような環境で既存手法を適用すると，多くの場合で破滅的忘却が発生し，ロット生産管理されている集積回路の製造現場では実用性に乏しいことが問題として挙げられる．この問題を克服するために，継続学習を利用する新しい継続学習ベースの集積回路テストを提案する． | |||||||
| EDGAR ANAROSSI | M, 1回目発表 | 知能システム制御 | 杉本 謙二☆, | 和田 隆広, | 松原 崇充 | ||
|
title: *** Robotic Flow Manipulation *** abstract: *** Following the current development in robotics research in the manipulation of non-linear tasks, modeling and controlling a high-dimensional non-linear task such as flow control still remains a challenge. While recently modeling a flow dynamics is possible using some deep learning methods, it requires a lot of resources to be able to produce a good result. Dynamic Mode Decomposition (DMD) is another data-driven method that could be used to approximate a dynamical system. While the performance of DMD itself is not as good as the deep learning approaches, DMD can still get a good enough representation/characteristics of a dynamical system given a very limited amount of data. Although we couldn't directly use those characteristics as is, we could utilize Metric Learning to identify the latent space in which the characteristics between different states can be measured. In this research we will explore how to apply Metric Learning on DMD's characteristics with the final goal of developing a robot which is able to control those characteristics on demand. *** language of the presentation: *** English *** | |||||||
| ZHU LINGWEI | D, 中間発表 | 知能システム制御 | 杉本 謙二☆, | 和田 隆広, | 松原 崇充 | ||
|
title: *** Scalable Safe Reinforcement Learning and Its Applications ***
abstract: *** Recently reinforcement learning (RL) has achieved impressive successes across a wide range of domains such as obtaining super-human level performance on video games and the game of Go. However, the applications of RL has been largely limited to games and robotics in laboratory. A crucial reason is that there is no safety ensuring mechanism in RL, which poses a great challenge to applying RL to real-world applications since the agent or its environment might be damaged during the trial-and-error learning process. In this talk I recap the first half of my PhD research which focuses on both novel theory and applications in safe RL, with an emphasis specially on large-scale problems such as chemical plant and robotics. *** language of the presentation: *** English *** 発表題目: *** この部分を発表題目に *** 発表概要: *** この部分を発表概要に *** | |||||||
| 清水 達也 | D, 中間発表 | 知能システム制御 | 杉本 謙二, | 浦岡 行治, | 小林 泰介 | ||
|
title: Higher Functionality of Home Appliances using Reinforcement Learning
abstract: Since home appliances are closely related to daily life, the pursuit of performance and functionality such as noise reduction and energy saving is being continued. In addition, white goods, which have a high penetration rate worldwide, need to work on energy saving from the viewpoint of contributing to environmental problems. However, since there is a concern that the product cost will increase in order to improve at the hardware level, it is necessary to improve the functionality and performance with the existing limited hardware and computational resources. Therefore, we apply reinforcement learning, which is one of optimization technology, to such home appliances to improve performance and functionality. We have developed a control algorithm that uses reinforcement learning to reduce noise and vibration in the washing machine, and achieved good effect. The proposed algorithm and experimental results will be explained. language of the presentation: Japanese 発表題目: 強化学習を用いた家電製品の高機能化 発表概要: 家電製品は生活に密接しているからこそ静音化や省エネ化などユーザ視点での性能追求が続けられている． また、世界的に普及率が高い白物家電は、環境問題への貢献という観点でも省エネ等に取り組む必要がある. しかしながら、ハードウェアレベルで改善するには製品コストの増加が懸念されるため，既存の限られたハードウェアおよび計算資源で高機能化や性能改善が求められる. そこで，最適化技術の一種である強化学習をこのような家電製品に適用して高機能化や性能改善を目指す． 洗濯機の低騒音・低振動化を実現する強化学習を用いた制御アルゴリズムを開発し、顕著な効果を実現した． 提案したアルゴリズムと実験結果について説明する。 | |||||||
| 嶋田 萌 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元☆, | 浦岡行治（物質科学）, | 片平真史（客員教授）, | 石濱直樹（客員准教授）, | 高井利憲（客員准教授） |
|
title: Investigation of safety methods to prevent collisions and the creation of new debris during the space debris disposal phase
abstract: Space debris, which is unwanted man-made material in space orbit, is a general term for debris generated by explosions and collisions such as failed satellites and rocket upper stages. This study systematically compares and examines a number of ongoing or proposed debris removal projects. Specifically, we will compare and verify points to be considered in the process of (1) not generating new debris due to factors such as collisions during the processing of space debris, and (2) not posing a threat to other spacecraft due to factors such as debris release during debris removal machine operations. In parallel, the safety of the system will also be verified from the point of view of system assurance, focusing on the arguments used to explain the system to third parties. Using the system assurance and safety arguments of several debris handling satellites, a comparative validation of debris removal projects will be carried out. In the future, after the comparative verification, the advantages and disadvantages will be clarified in terms of safety assurance. Our aim is to be able to deal with not only possible events but also unpredictable events. language of the presentation:Japanese 発表題目: 宇宙機システムに対する安全性を考慮したアーキテクチャ評価手法の提案 発表概要: 宇宙軌道上の不要な人工物であるスペースデブリは,故障した人工衛星やロケットの上段などの爆発や衝突によって発生する破片などの総称である. 2007年に中国が軌道破壊実験を行って以来,その問題の深刻さは研究者の手に余るものとなっている.人類の宇宙利用が拡大するにつれ,デブリ天体の数も増加し,宇宙の安定した利用を脅かしている. 現在,そして未来に向けて,悪化した軌道環境では,デブリの発生を防ぐための対策だけでは不十分で,デブリ被害に対する信頼性・安全性を積極的に確保する必要がある. そこで,本研究では,現在進行中または提案されているいくつかのデブリ除去プロジェクトを体系的に比較・検証していく. また,これと並行して,システムを第三者に説明する際の論拠を中心としたシステムアシュアランスの観点からも,システムの安全性を検証する. 複数のデブリ処理衛星のシステム保証と安全性の論拠を用いて、デブリ除去プロジェクトの比較検証を行っていきたい. 将来的には,比較検証の後,安全性確保の観点からメリット・デメリットを明確にする.想定される事象だけでなく,想定できない事象にも対応できることを目指す. | |||||||