| TAO XIANGKUN | M, 1回目発表 | 光メディアインタフェース | 向川 康博, 加藤 博一, 舩冨 卓哉, 久保 尋之, 田中 賢一郎 |
|
title:
Towards the advanced dual videography.
abstract: Dual photography is a technique which renders reciprocal images with an exchanged viewpoint in a projector-camera (ProCam) system. However, it applies only to static scenes and the pre-processing is quite time-consuming. While another method named dual videography allows us to capture reciprocal videos in real-time via a synchronized ProCam system. In this study, I investigate further on dual videography in order to obtain a better understanding of its properties and limitations. In this presentation, I will show the key results of the experiments I performed and discuss the potential improvement and application of dual videography. language of the presentation: English | |||
| 西山 輝 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 岡田 実, 藤川 和利, Youngwoo Kim |
|
Title:Fundamental Study on Influence of Intentional Electromagnetic Interference on IC Communication abstract: In the field of intentional electromagnetic interference (IEMI), there has been a focus on the threats that degrade the performance of a target device by using high-power EM wave. In addition, it is reported tat the threats of IEMI using several volts, which cause device malfunction without disrupting ICs, and extracting secret keys used for encryption stored inside the IC. For this threat, IEMI is applied to the calculation process of the internal IC. However, when IEMI is applied not to the calculation but communication process between ICs, it may modify not only the transmitted data but also the commands and causes to impair the confidentiality and integrity. Therefore, in this study, we examine the influence of IEMI on transmitted data between ICs by using the proposed environment that can evaluate error occurrence on transmitted data. language of the presentation:Japanese | |||
| 平尾 歩 | M, 1回目発表 | 数理情報学 | 池田 和司, 佐藤 嘉伸, 吉本 潤一郎, 久保 孝富, 福嶋 誠 |
|
title: Detection of depression relapse based on longitudinal life-log data
abstract: Depression causes loss of quality of life. It is estimated that the loss will continuously increase for the next 20 years. According to The Institute for Health Metrics and Evaluation, depression disorder increased from 1990 to 2017 in global trends in disability. This implies that there has been little improvement for the issue for a long time; therefore, we need to consider treatment about depression disorder. Depression is a disease that tends to recur after treatment, and it was reported that early treatment is important in depression because patients with early treatment or intervention have a better prognosis. In this study, we aim to predict the relapse of remitted patients suffering from depression based on the life-log. language of the presentation: Japanese | |||
| 福嶋 章悟 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 岡田 実, 安本 慶一, 藤本 大介 |
|
title: Proposal of evaluation method for information leakage by considering sampling timing when measuring radiated electromagnetic wave
Abstract: There have been reports of threats that measure electromagnetic waves emitted unintentionally from electronic devices and acquire information. Traditionally, these threats have been limited to the military and diplomacy fields, but the threats have spread to consumer electronics due to the lower prices of instruments and faster computers. For this reason, it is considered that risk evaluation will be required in the near future, for devices that have not been subject to threats. Therefore, in this study, we propose a technique to improve the acquisition of information in consideration of the sampling timing of the electromagnetic waves radiated from the consumer electronics, and establish a technology to evaluate the risk of information leakage through electromagnetic waves even with inexpensive equipment. language of the presentation: Japanese | |||
| HONG SEONGEON | M, 1回目発表 | 情報セキュリティ工学 | 林 優一, 藤川 和利, 安本 慶一, Youngwoo Kim |
|
title:Proposal of a Novel LASER Fault Injection Attack against LED on IoT Devices and Countermeasures in order to Prevent the Attacks
abstract: LASER fault injection attacks are one of well-known hardware security threats. Conventional LASER fault injection attacks have some limitations such as only focusing on crypto moduels, a possibility of leaving traces due to the high power of LASER causing permanent damages, and requirements of professinal knowledge and expensive equipments to perform the attack. Therefore, it was considered a rather difficult attack to perform for ordinary attackers. However, a recent study suggested a possibility of LASER fault injection attack by simply irradiating LASER against LED. By using a characteristic of LED, researchers changed a value of transistor to arbitrary values. However, in this research, they did not focus on actual devices. Thus, to confirm whether the above threat is feasible even in actual devices, I propose a new LASER fault injection attack against actual devices. For the experiment, a LASER(SLP-836A-37) will be irradiated against a LED which is attatched on Arduino UNO in order to change the value of Transistor(TO-236AB). If it turns out that the new LASER fault injection attack is effective to retrieve desired information, it could bring big threats for many actual devices which are often used in daily life. So, I propose a concept of countermeasures against this kind of threat. language of the presentation: English | |||
| 髙下 大貴 | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦, 松本 健一 |
|
title : Search for new ligands by semi supervised MHG-VAE
abstract : The goal of drug design is to identify novel molecules that have certain desirable properties. However, optimization in molecular space is extremely challenging, because the search space is large, discrete, and unstructured. For example, Gomez-Bombarelli et al. (2018) used variational autoencoder (VAE) using SMILES character strings as input to map a discrete molecular structure space to a low-dimensional continuous latent space, and optimized the molecular structure through joint training with a property score predictor from a latent vector. However, when we want to search ligand molecules, there is little label data. And, Using SMILES as a compound descriptor can cause molecular decoding errors. Thus, we suggest an effective model that solve the above two problems and generate a desirable ligand molecule from the continuous latent space. references : Rafael Gomez-Bombarelli, Jennifer N. Wei, David Duvenaud: “Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules”, ACS Cent. Sci. p268 - 276, 2018. language of the presentation : Japanese | |||
| 小山 千尋 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 小笠原 司, 林 優一, 妙中 雄三 |
|
title: Recognizing the ROS attack vector and potential threat
bstract: Japanese cabinet office advocates "Society5.0", which is defined at "A human-centered society that balances economic advancement with the resolution of social problems by a system that highly integrates cyberspace and physical space." In that society, robot is the device of reflecting the advanced calculation result to physical space. However Robot Operating System(ROS) does not does provide any some security feature such as authentication. It allows an attacker to make a robot do the unintentional operation. Denial of Service(DoS) attack and data injection are effective measures to ROS. In this presentation, we focus on the attack vector and potential threat on ROS. language of the presentation: Japanese 発表題目: ROSのに対する攻撃ベクターと潜在的脅威の検討 発表概要: 内閣府が提唱している"Society5.0"ではサイバー空間と物理空間が高度に融合し、経済発展と社会的問題解決の両立を目標としている。その中で、ロボットは情報収集・分析結果を物理空間に反映するデバイスに位置している。しかし、多くのロボットで使用されている開発用プラットフォームROSには認証などのセキュリティ機能が存在しないため、攻撃者が意図しない動作を誘発させる可能性がある。 情報セキュリティ脅威に挙げられるDoS攻撃や不正なデータ注入はROSに対しても有効であることがわかっている。 そこで本研究では、ROSの攻撃ベクターと潜在的脅威及び今後の研究実施計画について述べる。 | |||
| 東 佑樹 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 作村 諭一(BS), Sakriani Sakti |
|
title: Automatic Speech Recognition Error Recovery with BERT
abstract: In recent years, automatic speech recognition (ASR) has made remarkable progress and has come to be called "as high performance as humans." However, ASR systems still often make unrecoverable errors, especially when some noises are mixed with the input speech utterances. To address this problem, in this study, we propose a method to recover ASR recognition errors using BERT. BERT, which stands for Bidirectional Encoder Representations from Transformers, is a language understanding model that bi-directionally (left-to-right and right-to-left) learns the contextual relations between words in a text by masking out some of the tokens. Here, we treat ASR errors as the masked tokens and utilize BERT to recover the complete sentences. In this presentation, we will show our preliminary results in the noise-free condition, to investigate the possibility of applying BERT to recover erroneous words in the output of ASR. In the future, we will continue to improve ASR output given noisy speech data. language of the presentation: Japanese 発表題目: BERTを用いた自動音声認識の誤り訂正 発表概要: 近年、自動音声認識(ASR)はめざましい発展を遂げ、「人間と同程度の性能」と言われるまでになった。しかし、入力音声に雑音が混じってしまうと、依然として深刻な認識エラーを引き起こしてしまう。そこで、本研究ではBERTを用いてASRの認識誤り箇所を復元する手法を提案する。BERTは、Bidirectional Encoder Representations from Transformersの略で、テキストの一部をマスクすることで、単語間の文脈関係を双方向に学習する言語理解モデルである。ここでは、ASRによるエラーをマスクされたトークンとして扱い、BERTを使用して文章を復元させる。本発表では、ASRの出力で誤った単語を回復するためにBERTを適用する可能性を調査するために、ノイズのない状態での予備実験の結果を示す。 | |||
| 槇野 大樹 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸, 別所 康全(BS), 大竹 義人, スーフィー マーゼン, 日朝 祐太 |
|
title: Construction of statistical model of cup and stem placement by quantitative analysis of pre- and post-operative CT images for surgical planning in total hip arthroplasty
abstract: In total hip arthroplasty (THA), surgeons need to plan an appropriate implant types, sizes, positions and orientations, considering changes in the range of motion, leg length, bone and implant stability, and load balance. In one of our previous work, we built a statistical shape model using preoperative CT images and surgical plans as the training data, and implemented an automatic surgical planning system. However, the problem with this method is that postoperative information was not used for the model construction. In THA, the surgery may not be performed perfectly as planned due to several reasons including changes in the size and insertion position of the implant and misplacement due to unintended surgical operation. Therefore, in this research, we add postoperative information to the construction of the conventional statistical model, and create a statistical shape model that performs surgical planning considering the postoperative CT evaluation of the surgeries performed in the past. Then, by comparing the surgical plan created from the preoperative CT image with the constructed model with the postoperative CT image, quantitative evaluation is made on items such as size, position difference, and rotation difference. language of the presentation: Japanese | |||
| 佐山 大智 | M, 1回目発表 | ネットワークシステム学 | 岡田 実, 林 優一, 東野 武史, Dong Duong Thang |
|
title:Leaked Magnetic Fild Reducation for MIMO Wireless Power Transmission by Source Voltage Phase Control
abstract: reless power transmission (WPT) can be applied to various electric devices from smartphones to electric vehicles and has been expected as an innovative technology for next generation. Basically, the transferable power of a WPT system of one transmit coil and one receive coil is limited due to physical limitations of the coil materials and the power electronic devices. Therefore, multiple-input multiple-output (MIMO)WPT system has been recently studied in order to improve the transferable power by delivering multiple power streams simultaneously. In MIMO system, the power streams influence each other because of the effect of cross-coupling, and therefore it is difficult to provide constant voltage to the loads when the load resistances and source voltages vary during the charging process. In order to solve this problem, in the previous study, the constant voltage and isolation scheme of two-stream WPT have been proposed by adding the compensation circuit. When the two streams are isolated, adjusting the voltage phase of one stream might not affect output voltage of the other so much. This open a new possibility to reduce the magnetic field leaked to surrounding space by adjusting the phase difference of the sources in destructive manner. Of course, in such a case, the power transfer efficiency may deteriorate. In this work, we will optimize the phase difference of the sources to significantly suppress the leaked magnetic field while maintaining a high power transfer efficiency. language of the presentation: Japanese | |||
| 馬渕 俊弥 | 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:State Estimation of Flexible Spine in Quadrupedal Robot
abstract:Bio-inspired quadruped robots are desired to be used especially on disaster scenes since their capability to move on almost of landmass. Existing quadrupedal robots, however, only actuate their legs, although animals actively utilize their spine. Therefore, this study develops the quadrupedal robot with a flexible and active spine to improve its agility. In such a complicated robot, however, a part of the robot’s state is no longer observable, and therefore, the state estimation method without any prior knowledge of the complicated robot is required. To this end, a deep-learning-based method, which extracts the latent spaces of the observable and unobservable states, respectively, and reveals their relationship, is proposed. Using it, the unobservable state can be estimated from the observable state. In a toy problem for restoring missing images, the proposed method succeeded in restoring them from remaining images. language of the presentation:Japanese | |||