| 辻 茉佑香 | 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, 吉久保 尋之，田中 賢一郎 |
|
title: The interaction system with a non-device plane using the synchronous delay of a projector camera system
abstract: The laser scanning projector is capable of focus-free image projection, so an application for converting a projection surface into an interactive screen is expected. However, when we implement an interaction in which non-sensing function is not mounted on screen surface, we should require a special additional device such as a depth camera to judge if an object such as a finger touches with the surface. In this study, I try to develop an interaction system with a non-device plane by making a delay in the synchronization of the projector camera system. I shoot only the reflected light of an object slightly in front of the projection plane and judge the contact with the projection plane. I'll show the result of the collected image in front of the projection plane and detection image. language of the presentation: Japanese 発表題目: プロジェクタ・カメラシステムの同期遅延を用いた非デバイス平面とのインタラクションシステムの開発 発表概要: レーザー走査プロジェクタは、フォーカスフリーの映像投影が可能であることから、投影面をインタラクティブな装置へ転用するアプリケーションが期待される。しかし、接触面にセンシング機能が搭載されていないインタラクションを実装する際には、面に指などのオブジェクトが接しているかを判断するためにdepthカメラなどの特殊な追加装置が必要となる。ここで、本研究ではプロジェクタ・カメラシステムの同期に遅延を生じさせ、投影面よりわずかに手前に存在する物体の反射光のみを選択的に撮影することで、非デバイス平面とのインタラクションシステムの開発を試みる。実際に投影面の手前を選択的に撮影し、物体を検出した結果を示す。 | |||
| 小林 恒輝 | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 笠原 正治, 小蔵 正輝 |
|
title: Deep learning-based output feedback stabilization
abstract:Static output feedback (SOF) problem is important problem in control engineering. We specifically present a datadriven methodology for this problem by using deep learning techniques. We unfold the system signal flow to obtain a feedforward signal flow graph, which we regard as a neural network. Numerical examples are presented to show that our methodology can achieve a stabilization. language of the presentation: Japanese | |||
| 芳澤 健太 | 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: Deep Reinforcement Learning with Feedforward and Feedback Policies
abstract: In recent years, deep reinforcement learning (DRL) technology has been developed dramatically, and to date, it has the ability to beat humans with board games. One of the next frontiers for DRL is robotics for complicated tasks, such as door opening and bulk picking tasks. However, many of these robots act relatively slowly because, in reinforcement learning, data acquired by sensors, which cause certain delays from measurement and communication, is necessary for the robot to decide the optimal action to interact with environment. We therefore propose a new DRL algorithm with feedforward and feedback policies. The feedforward policy additionally given would compensate the feedback policy originally given by deciding the optimal action without sensor data, only with history of the robot’s behavior. language of the presentation: Japanese | |||
| 矢野 史剛 | M, 1回目発表 | 光メディアインタフェース | 向川 康博, 中村 哲, 舩冨 卓哉, 吉久保 尋之, 田中 賢一郎 |
|
title: A study on assisting to draw intermediate frames using 3D-model-trained neural network
abstract: In animation production, inbetweening is time-consuming task. Previous research shows inbetweening automation technology, however it can not apply dynamic movement frames. Additionally animations almost have dynamic frames, for the reason, it is necessary for assisting inbetweening technology to apply dynamic movement frames. To apply dynamic movement inbetweening, we propose an assisting inbetweening method using 3D-model-trained neural network. This method uses 3D-model to facilitate predicting an intermediate frame from key frames. To discuss the possibility of our method, we conducted an experiment testing the fundamental idea of our proposed method under the simple condition. In this presentation, we report the result of the experiment that showed the possibility of our method. language of the presentation: Japanese | |||
| 松本 彩香 | M, 2回目発表 | インタラクティブメディア設計学 | 加藤 博一, 小笠原 司, 神原 誠之, Alexander Plopski, 藤本雄一郎 |
|
title: Supporting Autonomy of Elderly in Daily Life Activities using Assistive Technologies
abstract: Sustained self-support of the elderly in their daily life activities is key in an aging society. However, the decline of their physical and mental functions makes this difficult. We investigated these aspects in two situations. First, in terms of the physical problem, we focused on the use of a robot arm, which supports upper-body impairment. In this situation, the user sometimes cannot see the target object that the user wants to grab due to robot arm size. We implemented a system using diminished reality (DR) technology that creates an illusion of making the robot arm disappear. Then, we evaluated its effect with 36 participants. Our results revealed that when the target object was occluded by the robot arm, an interface with DR could reduce more physical load over an interface without it. On the other hand, when the target object was not occluded, using the interface with DR contributed more cognitive load and was less preferred. Second, mental function problems such as memory loss and low motivation inhibit the elderly in maintaining daily life activities by themselves. To approach these two main causes, we suggested a reminder system with a virtual character. Communication with caregivers or family is the best solution to motivate the elderly, however, this is not always accessible. In the absence of an actual person, our system provides a virtual character for the elderly to develop a good relationship with. We hypothesized that this situation can increase the motivation of users in keeping a good routine of daily activities. After implementing the system, we will evaluate it in interviews with the elderly and caregivers. language of the presentation: Japanese | |||
| 中才 恵太朗 | D, 中間発表 | ソフトウェア工学 | 松本 健一, 安本 慶一, 石尾 隆, 畑 秀明, Raula G. Kula |
|
title: The Effect of Donation Badges on Response Time in the Eclipse Bug Reports
abstract: Donations play an important role in free/libre and open source software projects. Eclipse started its donation program Friends of Eclipse in December 2007. Currently, donors who contribute 35 USD or more qualify for the Friend of Eclipse status for one year, and are recognized on Bugzilla issue tracking system of a friend of Eclipse badge (donation badges). However, little is known about the impacts of donation badges. In particular, how badges might benefit the donors. Based on the framework of causal inference, we study how promptly developers respond to bug reports that have donation badges compared with bug reports without donation badges. The analysis revealed that the donation badges decreased response time by about two hours in median. We also conducted a survey on the Eclipse community contributors to ask how they perceive bug reports. Unexpectedly we find that contributors do not pay attention to donation badges compared to other factors in bug reports. Our findings suggest that contributors do not prioritize bug reports with donation badges consciously, but donation badges have implicit effects on response times. We consider the impact of donation badges can be explained as signals, where developers use technical and social information as signals to evaluate potential future contributions. language of the presentation: Japanese | |||
| 佐藤 拓馬 | 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:
Analysis of cell morphology by clustering of multiple stained histopathological images using deep learning
abstract: The computer aided diagnosis based on deep learning are not only helpful for classification, but also useful for feature extraction from given images, especially encoding image data into discrete representation helps us obtain new knowledge. Although most of the related studies are based on supervised learning that needs curated pathological knowledge, it is useful to extract characteristic features in the given images, using unsupervised machine learning in order to obtain new pathological findings. We applied cluster analysis using deep learning which is trained based on self-augmented training (SAT) and maximization of mutual information. It showed that it can classify those pathological images into discrete categories. Our findings can be transalated into practical use in clinical environments. Japanese | |||