| 三浦 明波 | 1661017: D, 中間発表 | 知能コミュニケーション 中村 哲, 松本 裕治, 須藤 克仁, Graham Neubig |
|
title: Syntactic Matching Methods in Pivot Translation
abstract: Pivot translation is a useful method for translating between languages with little or no parallel data by utilizing parallel data in an intermediate language such as English. A popular approach for pivot translation used in phrase-based or tree-based translation models combines source-pivot and pivot-target translation models into a source-target model, as known as triangulation. However, this combination is based on the constituent words’ surface forms and often produces incorrect source-target phrase pairs due to semantic ambiguity in the pivot language, and interlingual differences. This degrades translation accuracy. In this paper, we propose a approach for the triangulation using syntactic subtrees in the pivot language to distinguish pivot language words by their syntactic roles to avoid incorrect phrase combinations. Experimental results on the United Nations Parallel Corpus show the proposed method improves translation accuracy in all tested combinations of language. language of the presentation: Japanese | ||
| 河中 祥吾 | 1651037: 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, 中村 哲, 荒川 豊, 藤本 まなと |
|
title: Vehicle detection method using environmental sound collected from bicycle mounted smartphone
abstract: The number of bicycle users is increasing in recent years, and it is conceivable that the number of bicycle accidents will increase accordingly. In order to prevent a bicycle accident, it is necessary to travel a safe route for the bicycle. Traffic volume, vehicle type, vehicle speed, lateral distance are included as elements for judging the safety route. We propose a method to collect these information from the environmental sound collected by the microphone of the smartphone attached to the bicycle. language of the presentation: Japanese | ||
| 日高 真人 | 1651090: 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, 中村 哲, 荒川 豊, 諏訪 博彦 |
| title: A system for collecting and curating sightseeing information toward satisfactory tour plan creation abstract: Satisfactory tour planning entails providing tourists with tour plans that reflect both static information, such as tourists preferences and profiles; and dynamic information, such as crowded sightseeing spots and congested paths to these spots. Collecting, maintaining, and curating such dynamic information imposes a lot of time, monetary, and labor costs on those who manage tourist spots and the tourists themselves. In this paper, we propose a system that uses two novel mechanisms: (1) a participatory collection mechanism that efficiently and timely collects and updates both static and dynamic information about tourist spots and paths; and (2) a content curation mechanism that selects important items from the collected information and compiles them into tour plans. We conducted a pilot study where two categories of people, those who make plans before sightseeing and those who do not make plans, toured Kyoto with and without our system. As a result, we confirmed that our system is effective especially for those who make no plans before sightseeing. We also confirmed that sufficient static and dynamic information about spots can be collected and updated within a reasonable cost. language of the presentation: Japanese | ||
| 大野 大志 | 1651024: M, 2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, 久保 尋之, 田中 賢一郎 |
|
title: Practical BRDF Reconstruction from Real Object using Reconstructed Geometry of Multi-view Images
abstract: Three-dimensional (3D) shape reconstruction from multi-view images has become more popular and this is known as photogrammetry, but its appearance is not plausible enough. In this paper, we present a practical method to reconstruct the bidirectional reflectance distribution function (BRDF) from real object by using multi-view images for 3D reconstruction and a few additional images for obtaining the reflectance distribution. We use the reconstructed geometry from multi-view images for BRDF reconstruction, however, the reconstructed geometry and surface normal tend to contain artifacts and these artifacts considerably affect BRDF reconstruction. Therefore, we introduce assumptions that are the main contribution of our study to extract the reliable regions from the reconstructed geometry and clarify how to determine the region of optimal light and view directions. The results demonstrate that our method effectively acquires a plausible BRDF. language of the presentation:Japanese | ||
| 北野 和哉 | 1651041: M, 2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, 田中 賢一郎, 久保 尋之 |
|
title:Recovering Temporal PSF using ToF Camera with Delayed Light Emission
abstract:Recovering temporal PSFs are important for various applications, especially analyzing light transport. Some methods that use amplitude modulated continuous wave time-of-flight (ToF) cameras are proposed to recover temporal PSFs, where the resolution is several nanoseconds. Contrarily, we show in this paper that sub-nanosecond resolution can be achieved using pulsed ToF cameras and an additional circuit. A circuit is inserted before the illumination so that the emission delay can be controlled by sub-nanoseconds. From the observations of various delay settings, we recover temporal PSFs of the sub-nanosecond resolution. We confirm the effectiveness of our method via real-world experiments. language of the presentation: Japanese | ||
| 鈴木 大介 | 1651064: M, 2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, 田中 賢一郎, 久保 尋之 |
|
title: Analyzing fog effect with ToF camera
abstract: Every year, many car accidents occur due to fog. The visibility is reduced and it makes difficult to distinguish what happens on the road. This visibility deterioration is due to light scattering. In this study, we aim at producing defogged image based on amplitude modulated infrared Time of Flight (ToF) camera. First, we model a defogging method using only reflectance and distance of the object. Then, we analyze and remove fog effect in simulation. And finally, in future work, we will use experimental data to apply defogging. language of the presentation: Japanese 発表題目: ToFカメラにおける霧の影響解析 発表概要: 雨や霧の中では視界が悪く，自動車事故が発生する原因にもなっている．本研究ではこの問題を解決するべく，霧の中にある物体の鮮明化を目的とする．霧の中で視界が悪化する原因として，光の散乱があげられる．この影響を除去するために，振幅変調した近赤外光を照射するToFカメラに利用する．近赤外光と振幅変調の性質により，得られる情報量を増やし，鮮明化の手がかりをつかむ．まず，物体の反射率と距離の関係を利用して，観測値から霧による影響を解析し，その影響を除去する手法を考案した．シミュレーションによる確認を行った結果，霧の影響とそれを除去した物体の情報を得ることができた．今後は，実データに本手法を適用させ，鮮明化に繋げていく． | ||
| 宮田 明裕 | 1651104: M, 2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, 田中 賢一郎, 久保 尋之 |
|
title: Normal Estimation of a Translucent Object from Polarization States by Single Scattering
abstract: In computer vision field, many researchers work on 3D shape reconstruction from images. However, the problem of estimating 3D shape of translucent objects is difficult because they scatter incident light. In our research, we propose a new method of normal estimation using polarization, which is observed through a polarizer, as a geometric cue. Most objects in real world scatter incident light randomly. That makes problem complicated since the phenomenon arises multiple light path. Here we assume single scattering for simplicity and model scattering and refraction with respect to polarization states. We demonstrate that the proposed method can estimate the normal using synthetic data and also try normal estimation on real experiments. language of the presentation: Japanese | ||
| 野添 光 | 1651086: M, 2回目発表 | 数理情報学 池田 和司, 金谷 重彦, 吉本 潤一郎 |
|
title: Stochasticity Promotes Synchronized Gene Expression between Cells in Somite Segmentation
abstract: abstract: One of the most important life phenomenone, there is a somite segmentation which determins the basis of organism. The somite segmentation is confirmed in a growth process from zygote to organism. The somite segmentation divide cell population into somite, and generates forms spatial period. Previous research discovered that synchronization of clock gene hes7 is essential for the somite segmentation, and time period of hes7 is converted to spatial period. In addition, the somite segmentation has robust system toward space perturbation, but previous deterministic model don‘t reprocut this robust system. Therefore, we considered principle of robust system of the somite segmentation is stochastisity which can change a expression phase drametically, and modeled reaction path way as stochastic process. language of the presentation: Japanese | ||
| 石谷 智之 | 1651010: M, 2回目発表 | 数理情報学 池田 和司, 杉本 謙二, 吉本 潤一郎 |
|
title: Inverse principle component analysis: An approach to multi-view representation learning
abstract: In many applications, we encounter multi-view data in which objects of interest are described by different views or observers. Multi-view representation learning is a new direction in machine learning to discover an efficient representation of such data for capturing regularity and relationship among the objects. Although several conventional methods for the problem such as the canonical component analysis and the self-organizing maps are successful,they assume that all the objects are completely described in any view. However, multi-view data obtained by human participants through psychological experiments often include missing values and violate the assumption. To tackle this issue, we considered an inverse problem of the principle component analysis as an approach to partially observable multi-view representation learning. In addition, we derived a simple algorithm to solve this inverse problem using the singular value decomposition. Finally, we discussed the basic properties of the algorithm based on our numerical experiments. language of the presentation: Japanese | ||
| 児矢野 晋太 | 1651049: M, 2回目発表 | 数理情報学 池田 和司, 安本 慶一, 吉本 潤一郎 |
|
title: Online Portfolio Selection Based on the Sentiments in Stock Microblogs
abstract: Online portfolio selection is an application of online learning in the machine learning literature to the financial problem of portfolio selection. The objective of online portfolio selection is to maximize the cumulative return over sequential multiple periods, where two types of online portfolio selection approaches have been proposed, FollowtheWinner and FollowtheLoser. Although the former approaches were well studied so far, the latter approaches have been found to outperform the former empirically. Thus, we propose a new type of portfolio strategy combining Follow-the-Winner and Follow-the-Loser by applying a semisupervised learning method to the posts in stock microblogs. In microblogs, each stock has a thread of posts, some of which are associated with sentiment such as bullish or bearish. Our method estimates the missing sentiments in a supervised learning manner and uses them to construct portfolio. Our approach is to exploit not only reliable sentiments, but also less reliable and even false ones. language of the presentation: Japanese | ||