| 岩口 尭史 | D, 中間発表 | 光メディアインタフェース | 向川 康博, 佐藤 嘉伸, 舩冨 卓哉, 久保 尋之, 田中 賢一郎 |
|
title: Acquisition of indirect light transport for scene understanding
abstract: Light transport describes where the light from source reach on the camera sensor through the scene. A practical difficulty of the acquisition of the light transport is that is low signal to noise ratio due to global illumination. We propose an acquisition method by both modulating patterns of projector and controlling exposure mask at the same time. We show our method is able to measure 4D light transport with less noise in the presence of global illumination. language of the presentation: Japanese | |||
| ANTONIO VICTOR ANDREW ASUNCION | D, 中間発表 | 計算システムズ生物学 | 金谷 重彦, 佐藤 嘉伸, MD.ALTAF-UL-AMIN, 小野 直亮 |
|
Title: Connecting Prostate Adenocarcinoma Gene Expression to Image Features extracted from Deep Convolutional Networks
Abstract: A number of advances involving convolutional neural network architectures have been developed and implemented for computer-aided diagnosis. These studies have demonstrated that extracting morphological features from histological images can be effective for classification of subtypes of various diseases, especially cancer. On the other hand, varying types of gene expression or mutation data have been a rich and ubiquitous resource for studies involving cancer prognosis, survival, and others. The aim of this study is to establish a linear model to determine the correlation between the features extracted from histological images through CNNs and their corresponding gene expression data. Testing performance is at 92%. Initial implementation also displays a similar performance with taking the mean tumor probability from the softmax layer of Inception-v3.Moreover, initial implementation suggests that the performance of the convolutional classifier fairs similarly to the mean probability taken from the tiles of each whole slide. We also introduce an analysis of decile threshold to determine if some threshold better classifies the images. Further analysis shows that the convolutional classifier amplifies the mean probabilities gathered from the Inception architecture. The analysis following the results of this model serves as a good starting point for outcome and prognostic predictions. Developments in the starting CNN architecture (Inception-v4, et al.) and the convolutional classifier can prove beneficial for connecting the image clusters formed by the overlays and those formed by gene expression. Language of Presentation: English | |||
| FAN JIAYING | M, 2回目発表 | モバイルコンピューティング | 伊藤 実, 飯田 元, 柴田 直樹, 川上 朋也 |
|
title: Consideration for a Within-Subject Study Comparing Effectiveness of Cameras with Different Field of Views in Remote Collaboration
abstract: As mobile communication devices and high definition cameras has been used extensively, remote collaboration can be supported in a lot of scenes in our daily life, like remote surgery, machine repairing, online learning. Except for voice, cameras also play an important role in view sharing in remote collaboration activities, especially with the emerging of panorama camera, more working environment information can be provided to distributed partners. Even though the cameras always being used in life recording and workspace sharing, their impact on remote collaboration is still unknown to us. To get a better understanding of effectiveness of different field of view in workspace sharing, while also in remote physical collaborative behavior, process, perception and result performance. We conduct a within-subject study contains two tasks and three technology methods, which tasks are setting to Lego assembly with detailed manual or not, the technical methods including voice, cameras and panorama cameras. We hope the findings from this study could offer design and research implications, underlining the importance of wide angle view of workspace in remote collaborations. language of the presentation: Japanese | |||
| HOANG DAI LONG | D, 中間発表 | コンピューティング・アーキテクチャ | 中島 康彦, 岡田 実, 中田 尚, TRAN THI HONG, 張 任遠 |
|
Title: Complexity Reduction and Security Enhancement for 802.11ah PHY Transceiver
Abstract: In the Internet of Things (IoT) applications where things are connected and exchanged date wirelessly, security places an important role. In addition, IoT sensors have limitation on the power source and computing ability. Therefore, developing a wireless communication transceiver which is high-security, low power consumption, and low complexity for IoT sensors is significant. Recently, IEEE 802.11ah is a promising candidate for developing IoT sensor's wireless transceiver. However, the current version of 802.11ah standard does not focus on enhancing the system security. In this research, we solve that problem by proposing a low complexity and high-security encryption method at the physical layer (PHY) layer of 802.11ah. The evaluation results show that the implementation of our encryption does not affect to the bit error rate (BER) and packet error rate (PER) performance of the system while some conventional methods do. Language of the presentation: English | |||
| LE DINH DUNG | D, 中間発表 | コンピューティング・アーキテクチャ | 中島 康彦, 岡田 実, 中田 尚, TRAN THI HONG, 張 任遠 |
|
title: Efficient RLL-FEC Decoding Schemes for Flicker Mitigation in Visible Light Communication
abstract: Visible light communication (VLC) makes use of the outstanding advantages of light-emitting diodes (LEDs) for short-range optical wireless communication as well as illumination. In on-off keying (OOK) VLC systems, run-length limited (RLL) codes are widely used to avoid the perceived fluctuation in light intensity, referred to as flicker. Besides, forward error correction (FEC) schemes are essential to ensure a reliable data transmission and extend the distances. Since the dual use of FEC and RLL codes reduces the error correction performance and increase the complexity, effective decoding schemes should be taken into account. In this work, efficient RLL-FEC decoding schemes are proposed for improving the decoding performance of VLC systems. We carry out extensive experiments to evaluate the proposed decoders. Numerical results show competitive bit error rate (BER) performances. language of the presentation: English | |||
| TIAN ZHAOYIN | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 中村 哲, MD.ALTAF-UL-AMIN, 黄 銘 |
|
title: Arrhythmias Detection by Nonlinear Statistical Features for Electrocardiogram signal
abstract: The physiological homeostasis is maintained by the complex self-regulation based on multiple physiological parameters, such as body temperature and blood pressure. And the complex regulation can be expressed by the nonlinear features of the heart rate variability (HRV). The use of the HRV has been used to facilitate the diagnosis of patients with myocardial infarction. And further attempts to use the nonlinear parameters to analysis arrhythmias have been tried. Inspired by these researches, we tried to use the HRV-related parameters to capture distinct characters of different arrhythmias. We firstly tried to use Symbolic Coding and Shannon Entropy to distinguish multiple arrhythmias simultaneously, whose result showed that this parameter can’t distinguish the ventricular ectopic heartbeats (V) from atrial fibrillation (AF) and normal sinus rhythm (N). Secondly, we used the Multiscale Entropy (MultiEn) to get a more general description. By using the Refined Composite Multiscale Entropy, which is a modification of MultiEn, the distinct properties of the three arrhythmias (N, AF and V) can be captured and it suggest that the MultiEn may be useful in the detection of arrhythmias in a short and noisy electrocardiogram measurement. language of the presentation: English | |||
| NGUYEN QUYNH MAI | M, 1回目発表 | 知能コミュニケーション | 中村 哲 |
|
title: Sentence Generation based on Integration of Various Observation
abstract: Natural Language Generation (NLG) plays a crucial role in spoken dialogue system, it has a significant impact on a user’s impression of the system. As a consequence, NLG researchers are increasingly looking into questions beyond the technical challenges of enabling NLG interface. In my talk, I would like to present my research in the field of NLG which is a sentence generation based on integration of various observation. language of the presentation: English | |||
| SUMAILA NIGO | M, 1回目発表 | 知能コミュニケーション | 中村 哲☆ |
|
Title:Learning to spot automatons: a convolutional neural network to detect social bots on twitter Abstract: Social bots have been associated with endangering democracy, causing panic at the time of crisis, affecting the stock market, affecting the advance of public policy and many other aspects since the early days of social media. Thus, there is a pertinent need to develop defense mechanism against these entities. Previous research focus on user's behavioral patterns (e.g. activity patters), metadata and tweets' syntactical data. My current reasearch considers the semantical information from user's content (timeline tweets) to assess if a user is human or social bot. Language of the presentation: English | |||
| TRAN VAN HIEN | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Convolutional Neural Networks with Adaptive Weight Initialization for Relation Classification
abstract: In recent years, relation classification has gained much success by exploiting deep neural networks. For example, Convolutional Neural Networks (CNNs) have significantly improved the performance on the task by effectively capturing hidden features within sentences via continuous representations. In this research, we propose a CNN that adaptively initializes the weights of its convolutional filters, rather than randomly initializing them, and that pays special attention to the important text segment from the first entity to the second entity in each input sentence. Experiments on the SemEval-2010 Task 8 dataset show that our model achieves new state-of-the-art results without using any external lexical resources. language of the presentation: English | |||