| 田中 翔平 | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title: Incorporating Event Causality to Re-ranking for Conversational Dialogue Responses
abstract: In this presentation, we propose a novel method for re-ranking response candidates generated from conversational models by considering their event-causality relations. The method utilizes an event causality dictionary extracted from Web texts such as that "be stressed out" leads "relax to releave stress," for re-ranking of response candidates. Our re-ranking method gives higher scores to the candidates that have any event-causality relations to any utterances in a dialogue history. It is expected that our proposed method can select diverse, context-aware and logical responses. We calculated the coverage of the event causality dictionary automatically extracted from Web to get to know the ratio of response candidates that can be re-ranked by the proposed method. language of the presentation: Japanese 発表題目: 因果関係を用いた雑談対話応答のリランキング 発表概要: 本プレゼンテーションでは，ある2つの事態間に成立する因果関係（ストレスが溜まる → 発散する，など）を用いて，Encoder-Decoder などの対話モデルにより生成された雑談対話応答候補をリランキングする手法を提案する． 提案手法では，Encoder-Decoder モデルなどの既存の対話モデルに対話履歴を入力し，N-best 応答を生成する． 生成された N-best 応答候補のうち，入力として与えられた対話履歴中の発話との間に因果関係が認められるものには高いスコアを与え，N-best 応答候補をリランキングする． これにより，従来の対話モデルと比較して多様で文脈に沿った，論理的な応答が選択され易くなることが期待できる． 現在は，N-best 応答候補と対話履歴に対して因果関係辞書を適用し，どの程度の候補がリランキング可能かについての調査まで完了している． | |||
| 田中 雄基 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 |
|
title: Pelvis and Femur Segmentation of Diseased Hip Joint
abstract: Understanding of three-dimensinal shape of the diseased hip is important surgical planning of total hip arthroplasty. However, manual segmentation of the hip joint region is time-consuming and not cost-effective. Moreover, when we analyze pathological variation in large-scale database retrospectively, manually or semi-automatic segmentation is not a realistic option. Threfore, in this study, we propose a fully automatic segmentation of the diseased hip joint from CT images using CNN. language of the presentation: Japanese | |||
| 本村 駿乃介 | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title:
Detection of semantic anomalies in speech using EEG with Non-negative Matrix and Tensor Factorization
abstract: In human-computer interactions such as spoken dialogue systems, a computer often uses semantically wrong words (i.e. semantic anomalies) and a human recognizes such errors. Automatic detection of the mistakes can be applied to automatic correction of the words or evaluation of the system. N400, a well known component of time-locked electroencephalogram (EEG) signals, is the response of the brain related to semantic processing of the words and is observed by averaging the EEG signals of many trials. However, in order to apply to real-time communication, it is necessary to detect the semantic anomalies in the spoken sentences from the EEG signal of a single trial. For the detection from a single trial EEG, a feature extraction method of the EEG signal is very important because the EEG signals are data of high-dimension and low signal-to-noise ratio. Non-negative Matrix / Tensor Factorization (NMF/NTF) are the methods widely used for feature extraction of various kind of tasks (e.g. audio separation, document analysis) and they have been applied to the study of EEG. We propose the method to detect the semantic anomalies from a single trial EEG signal by extracting features of EEG using NMF/NTF. language of the presentation: Japanese | |||
| 田力 健人 | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Shape manipulation of deformable objects using tools
abstract: We propose an operation method to deform the deformable object in a desired shape when the robot cooks in the home environment. In the proposed method, visual information and tactile information are linked and analyzed. In the visual sense, we analyze the condition of the subject deformable object when the tool operates the target deformable object by RGB-D camera. In tactile sense, feedback control is performed by a force sensor attached to the robot hand. language of the presentation: Japanese | |||
| 帖佐 克己 | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title: Training Neural Machine Translation using Word Embedding-based Loss
abstract: In neural machine translation (NMT), the computational cost at the output layer increases with the size of the target-side vocabulary. Using a limited-size vocabulary instead may cause a significant decrease in translation quality. This trade-off is derived from a softmax-based loss function that handles in-dictionary words independently, in which word similarity is not considered. In this study, we propose a novel NMT loss function that includes word similarity in forms of distances in a word embedding space. The proposed loss function encourages an NMT decoder to generate words close to their references in the embedding space; this helps the decoder to choose similar acceptable words when the actual best candidates are not included in the vocabulary due to its size limitation. In experiments using ASPEC Japanese-to-English and IWSLT17 English-to-French data sets, the proposed method showed improvements against a standard NMT baseline in both datasets; especially with IWSLT17 En-Fr, it achieved up to +1.72 in BLEU and +1.99 in METEOR. When the target-side vocabulary was very limited to 1,000 words, the proposed method demonstrated a substantial gain, +1.72 in METEOR with ASPEC Ja-En. language of the presentation: Japanese | |||
| 辻 光顕 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元☆ |
|
title: Formal Safety Verification of Hybrid Systems with STAMP/STPA
abstract: Safety-critical systems, for example, autonomous vehicles and space systems, are required to be highly reliable and STAMP/STPA is a widely accepted hazard analysis method for those systems. However, it is known that manual executions of STAMP/STPA require huge effort and practice. Moreover, it is not established how STAMP/STPA can effectively be applied to a hybrid system, which contains both discrete and continuous behaviors. To address these issues, I propose a method to verify safety properties of a hybrid system using STAMP/STPA. In the proposed method, STAMP/STPA plays a role for generating safety constraints of a target system. For the verification part, I adopt formal method to show that the target system satisfies the generated safety constraints. I am planning to have some case studies of autonomous vehicles and space systems to confirm usefulness and effectiveness of the proposed method. In this presentation, I will show a current idea for a formal representation of hybrid systems in the context of hazard analysis with STAMP/STPA. language of the presentation: Japanese 発表題目: STAMP/STPAによるハイブリッドシステムの形式的な安全検証手法 発表概要: 自動運転車や人工衛星などに用いられるシステムには高い安全性が要求される。このようなシステムの設計に対して安全分析を行う手法として、近年STAMP/STPAが注目されている。しかし、従来の手法では手作業による分析の作業量が膨大であったり、離散的、及び連続的な振る舞いの両方を同時に持つハイブリッドシステムへの適用方法が確立していなかったりといった課題がある。そこで、本研究ではハイブリッドシステムに対する設計段階における安全性の検証を効率的に行う手法を提案する。具体的には、STAMP/STPAによってハイブリッドシステムの安全制約を導出し、形式検証手法を用いてモデル化およびシステムの安全性が満たされることに対する証明を行う。また、自動運転車や宇宙機の分野における事例をもとに手法の有効性を示すことを目指す。本発表では、提案手法の実現に向けて解決すべき課題を検討するとともに、STAMP/STPAにおけるハイブリッドシステムの表現法について現段階での方針案を示す。 | |||
| 梅津 𠮷雅 | M, 2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, 井上 美智子, 荒川 豊, 諏訪 博彦, 藤本 まなと |
|
title: Development of Energy Harvesters As A Sensor for Place Recognition on Wearables
abstract: A wearable based long-term lifelogging system is desirable for the purpose of reviewing and improving users’ own lifestyle habits. Energy harvesting (EH) is a promising means for realizing sustainable lifelogging. However, present EH technologies suffer from the instability of generated electricity caused by the change of environment, e.g., an output of a solar cell varies by its material, light intensity, and wavelength. In this paper, we leverage this instability of EH technologies for other purposes, in addition to an energy source. Specifically, we propose to reckon the variation of generated electricity as a sensor for recognizing “places” where the user stays, which are important information in the lifelogging system. First, we investigate the amount of generated electricity of selected energy harvesting elements in various environments. Second, we design a system called EHAAS (Energy Harvesters As A Sensor) where energy harvesting elements are used as a sensor. With EHAAS, we propose a place recognition method based on machine-learning and implement a prototype wearable system. Our prototype evaluation confirms that EHAAS achieves place recognition accuracy of 88.5% F-value for nine different indoor and outdoor places. This result is better than the result by existing sensors (3-axis Acc. & Brightness). language of the presentation: Japanese | |||
| 菅田 唯仁 | 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 |
|
title: Energy-harvest wattmeter for energy harvesting elements
abstract: The place recognition system based on generated electricity of energy harvest elements has been proposed. However, it uses an external battery for measuring the generated power. Toward the energy-harvesting of the whole system, we propose a energy-harvest wattmeter that can measure the generated electricity without a battery. language of the presentation: Japanese 発表題目: 環境発電による環境発電素子の発電量計測システムに関する研究 発表概要: 環境発電素子の発電量をセンサ値とみなし，場所や行動を認識できることが示されているが，発電量の計測に外付けバッテリーを必要としている．本研究では，システム全体のエナジーハーベスト化を目指し，環境発電素子の発電量を発電した電力を用いて計測するシステムを提案・実装する． | |||
| 出口 和樹 | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦 |
|
title:Correlation analysis of characteristics of lung cancer morphology and gene expression using convolution neural network
abstract:It is known that cancer occurs due to mutation in genes. However, it is not known which genes cause the phenotype of the cancer tissue. In this study, we will classify lung cancer tissue images using auto encoder, which is a kind of convolution neural network, in order to investigate which gene the phenotype of cancer tissue is derived. By using the feature value obtained by classification, which gene is the difference in expression system of lung cancer tissue is examined. language of the presentation:Japanese | |||
| 土井畑 禅 | M, 1回目発表 | 数理情報学 | 池田 和司☆ |
|
title: Neural control strategies underlying human-human joint action
abstract:Social and physical joint actions are characterized by the emergence of naturally occurring mutual cooperative relationships. The role in this cooperative relationship varies depending on various factors, and the mechanism of its appearance is not understood. If we can understand this mechanism, we can support learning of human movements between humans and humans, and we can improve interaction between humans and robots. In this study, we explore the collaborative work in which the role in cooperative relationship appears and investigate the mechanism of role occurrence. language of the presentation:Japanesea | |||
| 時末 尚悟 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 |
|
title: Musculotendinous modeling by using fiber tractography of high-resolution cryosectioned images
abstract: Musculotendinous junctions play an essential role in supporting human body by connecting between the bones and skeletal muscles. However, current musculoskeletal models do not represent the detailed structure of the junction. Therefore, in this research, we aimed at developing a modeling technique of the musculotendinous junction in high-resolution cryosectioned three-dimensional images. A region-of-interest including the junction was extracted from the image, and the junction was manually traced. Next, the junction was mathematically modeled by estimating the fiber directions based on structure tensors and applying a muscle fiber tractography approach developed in a previous study. The technique was applied to the junction connecting the gluteus maximus muscle and the iliotibial tract in a high-resolution image (voxel size: 0.1 mm) selected from the Visible Korean Human dataset. The derived junction model could clarify the boundaries among the constituent tissues. Our future work will focus on validating the proposed technique on junctions at different anatomical regions. language of the presentation: Japanese | |||
| 中尾 文香 | M, 1回目発表 | 情報セキュリティ工学 | 林 優一 |
|
title:Basic Study on Arbitrary Command Injection Threats against Actuators and Sensors inside IoT device
abstract:Along with the spread of IoT device, a wide variety of devices are being connected to the network. By transmitting the encrypted command to the IoT device connected via communication networks, the desired operation is executed. However, in general, the actuators and sensors connected to the IoT device do not have cryptographic functions. For that reason, the encrypted data received via the network is decrypted inside the device, and get action by sending an electric signal based on the obtained data. Therefore, it is difficult to guarantee the integrity of the signal transmitted to the actuator or the sensor, and the attacker may eavesdrop, injects or tamper with the electric signal representing the operation command. In this research, we construct an evaluation environment to investigate the possibility of injecting arbitrary commands by using EM waves to actuators and sensors and consider the feasibility of the attacks mentioned above using that environment. Moreover, we will discuss countermeasures. language of the presentation: Japanese | |||
| 永島 幹久 | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Development of assist device for knee osteoarthritis using movement estimation
abstract: Knee osteoarthritis is a disease which is caused by aging, obesity, etc. Generally, this disease causes pain when heel strike and start walking. However, in other researches they estimate walking cycle and heel strike, someone estimate start walking. So, we estimate start walking and develop assist devise with using insole sensor and IMU sensor. language of the presentation: Japanese 発表題目: 動き出し推定を用いた変形性膝関節症向けアシスト装置開発 発表概要: 変形性膝関節症は加齢や肥満などにより膝の軟骨がすり減り，痛みが生じる疾患である．一般的にこの疾患はかかとの接地時と動き出しのタイミングで痛みが生じる．多くの研究では歩行周期やヒールストライクのタイミングの推定は行われているが，動き出しの推定はあまり行われていない．この動き出しを推定し，痛みを軽減するアシスト装置の開発を行う | |||
| 仲田 勇太 | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Automatic summarization of scientific paper using citation information
abstract: Increasing and digitizing scientific papers raised the demand for automatic summarization. Extractive summarization are automatic summarization methods that have been studied for many years. In this presentation we will report about extractive summarization using cited sentences in scientific paper. language of the presentation: Japanese | |||
| 永谷 智貴 | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Safe control method of robot arm considering human contact
abstract: High robustness is required for the robot arm used in the home so as not to harm peoples and objects.Conventionally, safety has been secured by avoiding contact. On the other hand, large detours and unnecessary stops to avoid contact worsen work efficiency and lose the advantages of robot installation.In this research, in order to enable efficient operation with small detouring and minimum stop, by detecting the surrounding human beings and objects, by reducing the speed at the time of contact and impedance control, reducing the risk of contact, develop a method that keeps safety even if it occurs, and confirm compatibility of safety and efficiency. language of the presentation: Japanese 発表題目:人間との接触を考慮したロボットアームの安全な制御手法 発表概要: 家庭内で使われるロボットアームには、周囲の人間や物体に危害を加えないために高い安全性が求められる。従来、安全性は接触を避けることによって確保されてきたが、一方で接触を避けるための大きな迂回や不必要な停止は作業効率を悪化させ、ロボット導入の意義を失わせる。本研究では、小さな迂回や最低限の停止で効率的な動作が出来るよう、周囲の人間や物体を検出し、接触時の速度の緩和やインピーダンス制御により接触時の危険を減らすことで、接触が発生しても安全性が保たれる手法を開発し、安全性と効率の両立を確認する。 | |||
| 中野 萌士 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清 |
|
title:Gustatory manipulation interface cross-modal effect of vision on taste elicited with AR based real-time food appearance modulation
abstract: Appetite is one of the three greatest desires of human beings, which is indispensable for living, enriching people’s lives. The pleasure of being able to eat what you want to eat, such as your favorites and high-class ingredients, is invaluable. However, there are circumstances in which you cannot eat what you want to eat because of a variety of reasons such as dietary restrictions, food allergies, and religious restrictions. In this presentation, I propose a gustatory manipulation interface cross-modal effect of vision on taste elicited with AR based real-time food appearance modulation using 3Dmodel. language of the presentation: Japanese | |||
| 長濵 将太 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子 |
|
title: Graph Exploration with Myopic Luminous Robots
abstract: In this research, we aim to find out graph exploration algorithms for autonomous mobile robots. We assume that capability of robots is very weak. Thus this research can be applied to the system consisting of small robots such as microrobots or molecular robots. In this presentation, we propose two ring exploration algorithms. The algorithms, for which we require extending the visibility of robots a little, achieve exploration with a smaller number of robots than existing algorithms. language of the presentation: Japanese | |||