| 渡邉 大生 | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一 |
|
title: A Research to Solve Narrow FoV Problem for AR-HMD.
abstract: On general application based on Augmented Reality (AR), there is narrow Field of View (FoV) problem for Head Mounted Display (HMD). On the other hand, when target object to focus appear, human have characteristic that is the following of face delayed compare to gaze motion. That is one of human natural motions. However, the case that we consider like object assembly task as one AR application, the above natural motion is inhibited by narrowness of HMD. Therefore, I would like to survey to solve this problem. language of the presentation: Japanese 発表題目: ARにおけるHMDの狭視野角の解決に関する研究 発表概要: Augmented Reality (AR)を用いた一般的な応用では，しばしばHead Mounted Display (HMD)の狭さが問題として取り上げられる．一方で視覚に関する人間の自然な動きとして，人の視線が目標物体にフォーカスされた後に顔の正面が目標に追従するという特性がある．ここで物体組立を支援するようなARの応用を考えると，HMDのField of View (FoV)が狭い場合に上記の自然な動作が阻害されるという問題がある．そこで私はこの問題を解決する方法についての調査を行う． | |||
| BUTASLAC ISIDRO Ⅲ MENDOZA | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一 |
|
title: Enhancing Visual Perception: Freeform Depth of Field via Time Multiplexed Occlusion
abstract: Technology has rapidly advanced in the last few decades, but our growth as humans is capped by the current limitations of our human body. This study aims to modulate human visual perception to attain visual performance that lie beyond existing human capabilities. Our method is to make a "freeform depth of field", thereby controlling the amount of visual information a person can see. Through this modulation a person's visual perception will be enhanced. language of the presentation: English | |||
| CHANG CHAO-LING | M, 1回目発表 | 数理情報学 | 池田 和司 |
|
Title: Obstacle Detection from Driving Behaviors Using Maximum Mean Discrepancy and Spatial Clustering Abstract: Obstacles on lane or damage to roads often cause fatal car accidents. Monitoring the roads by equipping with many cameras can be a potential yet expensive solution. As an alternative, Ikeda et al., used driving behaviors to detect the obstacles on the road by formulating it as a change point detection problem. They used the nonparametric method Maximum Mean Discrepancy (MMD) to perform the detection. They demonstrated that the accuracy of detection increases when considering the driving behaviors from neighboring regions as well. However, it is unclear how the regions need to be clustered for optimal performance. In my work, I want to first apply a region clustering method to classify regions here in order to improve the accuracy of detection. Second, design an experiment to test the sensitivity of MMD score for different variables. With these two works, I hope to find the suitable clustering of regions and limitation of using MMD method. Language of the presentation: English | |||
| ZAVIALOV IGOR | M, 1回目発表 | 数理情報学 | 池田 和司 |
|
title: New framework for measuring and stress-testing the systemic risk of the financial system
abstract: My research focuses on the measuring the systemic risk of the banking system. The systemic risk is defined as the probability that the default of one institution will make other institutions default. The proposed framework will use set of unsupervised machine learning techniques such as Deep Learning and clustering algorithms for assessing and classification the systemic risk under various economic conditions. One of the parts contributing to the systemic risk value is a value of bank’s portfolio estimated by Value-at-Risk (VaR) measurement model where I am going to propose Markov Chain Monte Carlo (MCMC) sampling for VaR estimation and compare it with existing methods such as variance-covariance, historic and Monte Carlo simulation methods. language of the presentation: English | |||
| FITO WIGUNANTO HERMINAWAN | M, 1回目発表 | ネットワークシステム学 | 岡田 実 |
|
title: Subchannel Allocation for LTE Uplink in Radio over Fiber Mobile Fronthaul
abstract: Radio over Fiber (RoF) technology is hybrid technology which combine fiber-optics and radio wireless system. It can be promising for constructing LTEmobile front-haul link, however, transmission quality is suffered from intermodulation distortion (IMD) with frequency dependency when themultiband signals are transmitted. In current LTE system, subchannel resourcein uplink is allocated to mobile terminal (MT) so that its capacity is maximized referringfrequency response of air propagation channel. Above frequency dependency of IMD, however,is not taken account. In this research, subchannel allocation isoptimized by jointly considering frequency selective response of air propagation and frequency dependency of IMD in fiber optics transmission. language of the presentation: English | |||
| HU JIEYING | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦 |
|
title: Prediction of hERG Potassium Channels Affinity by Convolutional Network on graphs for Potential Drug Screening
abstract: Toxicity is a central issue in the development of new drugs, there are numbers of drugs failing in clinical trials or even need to be taken off of the market because of the toxic effects. Cardiotoxicity has become a leading cause for drug failure. Although the High-Throughput Screening HTS is available for testing the toxicity of a high number of chemicals, time spending on testing a compound several times at different concentration levels is not the best ideal way. CNN can be conducted on the chemical structural graphs directly and it has demonstrated its good performance on images recognition and classification in the last decade. In this research, we propose convolutional neural network to develop QSPR model for predicting hERG channel adverse cardiac effects for drug virtual screening and drug design at the early steps of the drug discovery process. language of the presentation: English | |||
| TULATHUM PATTARAPORN | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Tactile feedback for material identification
abstract: The sense of touch in a human being is a bit different than the other 4 senses. Unlike the other senses which are located at specific body parts, touch is a sense that is all over our body. In a robot, this sensing can enable a robot to infer properties of its surroundings, such as the material of an object. In this work, we explore how to improve tactile sensing ability in robot consider by the characteristics of materials. So, I will identify types of materials by using the tactile information from the combination of 2 sensors, OptoForce sensor and Thermistor sensor. language of the presentation: English | |||
| WANG ZIYU | M, 1回目発表 | ロボティクス | 小笠原 司 |
|
title: Robust Initialization for Monocular Visual-Inertial SLAM
abstract: Recently, there is a trend of assisting the monocular vision system with a low cost inertial measurement unit(IMU), since the two sensor are small, cheap low power consumption and able to complement each other. In this research, we propose a loosely-coupled method to provide accurate initial value to bootsrap the nonlinear optimization system and implement camera-imu extrinsic and time offset calibration automatically. language of the presentation: English | |||
| YU MIAO | M, 1回目発表 | インタラクティブメディア設計学 | 加藤 博一 |
|
title: Large-sized 3D Object Manipulation in Augmented Reality Environment for Urban Planning Tasks
abstract: With the development of Augmented Reality (AR) technology, more advanced AR devices are becoming commodity hardware available to the average user and feasible to use as a tool for 3D work. This will make AR applications more widely available in various scenarios. Interaction with objects for 3D work is essential in AR environment. In the past, a variety of interaction methods have been presented. However, there has been little research on the interaction with large-sized 3D object. In this research, we attempt to propose a method for natural, gesture based interaction with large-sized 3D objects in Augmented Reality environments and apply to the urban planning tasks. language of the presentation: English | |||
| CHOI JUNG WON | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Lexicon-based Deep Learning Model for Text Classification and Aspect Terms Clustering
abstract: With the development of the internet, classifying reviews has become an important task in Natural Language Processing (NLP). The Neural Networks model, using a joint framework between Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), has recently significantly improved performance. The review classification task has primarily been studied as separated topics, as either the task of distinguishing the positive and negative reviews, or a task of extracting aspect terms. We propose applying lexicon information to a single, joint CNN-RNN model to classify text polarity and clustering extracted aspect terms to analyze the reasons behind polarity. In this presentation, I will be introducing related research materials and our proposed model. language of the presentation: Japanese | |||
| ZHAO XIN | M, 1回目発表 | 自然言語処理学 | 松本 裕治 |
|
title: Entity alignment among Cross-lingual Knowledge Base
abstract: Knowledge bases are built in several different languages, achieving cross-lingual knowledge alignment will help people in constructing a coherent knowledge base, and assist machines in dealing with different expressions of entity relationships across diverse human languages. However, recent works only focus on two-graphs alignment, which cannot use the abundant information about other knowledge graphs. In this research, aligning entities among two more graphs efficiently is our concerns. language of the presentation: English | |||