| 中川 豊 | D, 中間発表 | ネットワークシステム学 | 岡田 実, | 藤川 和利, | 東野 武史, | Duong Quang Thang | |
|
title: Research on meteorological forecasts applying radio science abstract:Local meteorological disasters such as guerrilla rainstorms and tornadoes are caused by rapidly developing cumulonimbus clouds, but high-frequency and high-density observations commensurate with the development of cumulonimbus clouds are difficult with existing observation techniques. It is also difficult to make accurate predictions even with the latest supercomputers. In this research, we aim to improve the accuracy of weather forecasts for free by making full use of machine learning and deep learning based on open data obtained from existing systems such as GNSS. We will also conduct research and development for the utilization of disaster prevention systems in tropical regions where observation and prediction systems are vulnerable. language of the presentation: Japanese 発表題目: 電波科学を応用した気象予測の研究 発表概要: ゲリラ豪雨や竜巻などの局地的な気象災害は急激に発達する積乱雲からもたらされるが、既存の観測技術では積乱雲の発達に即した高頻度・高密度な観測が難しく、最新のスーパーコンピュータをもってしても早期予測が困難である。本研究ではGNSSなど既存のシステムから得られるオープンデータを基に機械学習、深層学習等を駆使してコストをかけずに気象予測の精度向上を目指す。さらに観測、予報体制の脆弱な熱帯地方への防災体制等の利活用に向けた研究開発を行う。 | |||||||
| WATANAKEESUNTORN WASSAPON | D, 中間発表 | ソフトウェア設計学 | 飯田 元, | 藤川 和利, | 市川 昊平, | 髙橋 慧智 | |
|
title: Massively Parallel Causal Inference of Whole Brain Dynamics at Single Neuron Resolution
abstract: Empirical Dynamic Modeling (EDM) is a nonlinear time series causal inference framework. The latest implementation of EDM, cppEDM, has only been used for small datasets due to computational cost. With the growth of data collection capabilities, there is a great need to identify causal relationships in large datasets. We present mpEDM, a parallel distributed implementation of EDM optimized for modern GPU-centric supercomputers. We improve the original algorithm to reduce redundant computation and optimize the implementation to fully utilize hardware resources such as GPUs and SIMD units. As a use case, we run mpEDM on AI Bridging Cloud Infrastructure (ABCI) using datasets of an entire animal brain sampled at single neuron resolution to identify dynamical causation patterns across the brain. mpEDM is 1,530x faster than cppEDM and a dataset containing 101,729 neuron was analyzed in 199 seconds on 512 nodes. This is the largest EDM causal inference achieved to date. language of the presentation: English | |||||||
| LI GUOQING | M, 1回目発表 | ソフトウェア設計学 | 飯田 元, | 太田 淳, | 藤川 和利, | 市川 昊平, | 高橋 慧智 |
|
Title: Kubevirt and the cost of containerizing VMs Abstract: KubeVirt is an add-on for Kubernetes to manage both containers and VMs in a unified manner. In KubeVirt, however, Libvirt, QEMU and all the VM processes run in Kubernetes pods and this may introduce some overhead. For instance, resource usage accounting or limitation done by the container runtime, as well as longer (disk and network) IO paths might slow down KubeVirt VMs. I will investigate the performance characteristics of VMs running in Kubernetes pods, using CPU, memory, disk and networking benchmarks and comparing that with plain KVM VMs. Furthermore, I will check the effect that tuning the configuration of the plain VMs has, for these benchmarks, and show whether it is possible to improve the virtualization performance for VMs running with KubeVirt. Language of the Presentation: English | |||||||
| NIU ZHAOFENG | D, 中間発表 | インタラクティブメディア設計学 | 加藤 博一, | 清川 清, | 神原 誠之, | 藤本 雄一郎 |
|
Title: The application of deep CNN models in 3D scene modeling Abstract: 3D scene modeling is an important key for many applications, such as 3D reconstruction and automatic driving. There are existing depth cameras which can obtain the depth information and build the 3D models. However, the depth cameras are expensive and most of them can only be used for indoor scenarios. To achieve 3D scene modeling for more general scenarios, two steps are considered: 1) use a deep CNN model to obtain estimated depth maps based on a single RGB image; 2) considering both depth noise and pose noise, fuse the depth maps to build the 3D model with a deep CNN model. Two steps are conducted separately in the first place, then will be combined together, so that the 3D scene modeling can be achieved from RGB images to a 3D model. Language of the Presentation: English | ||||||
| CHOTCHAICHARIN SETTHAWUT | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, | 加藤 博一, | 酒田 信親, | 磯山 直也 |
|
title: Compelling AR Earthquake Simulation with AR Screen Shaking
abstract: Many countries around the world suffer losses from earthquake disasters, most of which are inevitable. To reduce the injury of individuals, safety training is essential to raise people's preparedness. To conduct virtual training, previous work mostly uses virtual reality (VR) to mimic the real world, without considering augmented reality (AR). Our goal is to simulate earthquakes in a familiar environment helping users to take the simulation more seriously. We propose an augmented reality earthquake simulation using a video see-through VR headset. We also implement a novel AR screen shake technique, which simulates the forces applied to the user's head by shaking the entire view. This allows us to achieve a simulation experience of high realism. We run a user study where participants experienced an earthquake both in a VR scene and two AR scenes with and without the AR screen shake technique. Our results suggest that both AR scenes offer a more compelling experience compared to the VR scene. Furthermore, we found that the AR screen shake improved immediacy and was preferred by most participants. This showed us how virtual safety training can greatly benefit from an AR implementation, motivating us to further explore this approach for the case of earthquakes. language of the presentation: ENGLISH | ||||||
| 喜多山 湧也 | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, | 加藤 博一, | 酒田 信親, | 磯山 直也 |
|
title: Development and Evaluation of a DisplaySystem that Presents Information Only in thePeripheral Vision
abstract: It is well known that the peripheral vision information has various mental effects, such as evoking a sense of security and empathy.Whereas, most existing information presentation systems cover the central vision with a display, which is detrimental to the actual work.This study aims to construct a system that can take the effect at the right place without interfering with the central vision by providing displays only in the peripheral vision.Preliminary experiments were carried out using a prototype that assumed sympathy scenarios, and the effectiveness of this method was confirmed. language of the presentation: Japanese 発表題目: 周辺視野のみに情報提示するディスプレイシステムの開発と評価 発表概要: 周辺視野情報が安心感や共感の喚起など，様々な精神的効果を与える事が明らかになっている．一方，既存の情報提示システムの多くは中心視野もディスプレイで覆うもので，実作業の弊害になるという問題がある．本研究では周辺視野のみにディスプレイを備えることで，中心視野を阻害することなく適材適所でその効果を受けられるシステムの構築を目指す．試作機を用いて共感を与えるシナリオを想定した予備実験を実施し，本システムの有効性を確認した． | ||||||