| HOVHANNISYAN ANI | M, 2回目発表 | ソフトウェア工学 | 松本 健一, 飯田 元, 石尾 隆, Raula Gaikovina Kula |
|
Title: Facilitating Source Code Analysis: A Differential Approach to Reducing Warnings of Software Source Code Static Analysis Abstract: Software source code is getting larger and larger nowadays. And its maintenance is becoming difficult due to its large amount of source code. Currently, there are various recommended Automatic Static Analysis Tools (SAT) that help to reduce manual work of developers. However, most SAT provide too long list of warnings, which contain noisy warnings or aren't related to developers desired part of source code changes. Therefore, for facilitating source code analysis, this research proposes to apply differential analysis to the "changed" source code fragments instead of "all" source code. To apply this methodology research has developed a method-level extractor tool which parses source code and extracts only changed parts of source code. Then the tool passes newly generated source files to the Static Analysis Tool as an input. Research also performed 4 cases of experiments, that show the proposed differential approach to facilitate Source Code Static Analysis is effective, and the warnings of Software Source Code Static Analysis was reduced. Language of the presentation: English | |||
| HONG RUIXUN | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 佐藤 嘉伸, 黄 銘, MD.ALTAF-UL-AMIN, 小野 直亮 |
|
title: Anti-inflammatory Activity Prediction of Plant Secondary Metabolites Based on Machine Learning Models abstract: The therapeutic potential of plant secondary metabolites is currently a research target in the pursuit of novel anti-inflammatory drugs. In this study, I constructed four datasets composed of 174 anti-inflammatory metabolites as positive data, along with an equal number of different negative data samples. I represented the metabolites using ECFP (Extended-Connectivity Fingerprints) to utilize them as inputs for machine learning algorithms and compared ECFP with MACCS fingerprints. I evaluated the performance of various machine learning models, including DNNs, SVM, RF, and XGBoost, for predicting the anti-inflammatory activity of plant secondary metabolites. language of the presentation: English | |||
| 大杉 和寛 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘 |
|
title: Development of a Model of Computer Aided Diagnosis for Scoliosis and Pleural Thickening Using Chest X ray imaging
abstract: We aim to build models to assist physicians in diagnosing scoliosis and pleural thickening. Previous studies have demonstrated high classification accuracy using CNN-based convolutional models and Vision Transformer (ViT)-based models in deep learning of image tasks. In this study, we compare the classification accuracy of scoliosis and pleural thickening using chest radiographs. language of the presentation: Japanese 発表題目: 胸部X線写真を用いた側弯症及び胸膜肥厚の診断支援モデルの構築 発表概要: 医師の診断を支援するという目的のため側弯症及び胸膜肥厚の診断支援モデルの構築を目指している。これまでに画像タスクの深層学習においてCNNをベースにした畳み込みを用いたモデル及びVision Transformer (ViT)をベースにしたモデルで高い分類精度を出している。そこで本研究では胸部X線写真を使用し、これらのモデルで側弯症及び胸膜肥厚の分類精度を比較し,その結果を示す。 | |||
| WANG ZIHANG | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN |
|
title: Drug Repurposing for Non-Alcoholic Fatty Liver Disease Based On Relations Among Drugs,Disease and Genes abstract: As our understanding of the genetics driving non-alcoholic fatty liver disease (NAFLD) continues to evolve, we persistently strive for methodologies that can meaningfully apply this data within a clinical setting. Genome-wide association studies have flagged a multitude of genetic risk loci for NAFLD. The identification of these risk loci expands our understanding of the disease, but their clinical relevance remains limited. Consequently, we propose to harness this genetic information to identify potential NAFLD therapeutics that may target these loci, a process termed drug repurposing. The present research propose a highly efficient and cost-effective computational drug repositioning technique that not only focuses on the interplay between drugs and risk genes but extends to the interactions between drugs and all biological pathways that the risk genes involve. The ultimate aim of this study is to pinpoint candidate genes for NAFLD, unearth potential drug targets, and identify drugs that can be repositioned or developed specifically for NAFLD treatment. language of the presentation: English | |||
| CHUANG ZHONG | M, 1回目発表 | 数理情報学(コミュニケーション学) | 池田 和司, 岩田 具治, 田中 佑典 |
|
title: Meta-learning from Graphs in Heterogenous Attribute Spaces
abstract: The prediction of relationships between nodes in network data has achieved notable advancements, largely attributed to graph-based machine learning methods. However, these methods often rely on message exchange facilitated by known linkages in the training data. When there is an insufficient number of known linkages in the training data, building an effective model becomes challenging. In this research, we developed a meta-learning model capable of training on graphs in heterogeneous attribute spaces and making predictions on unknown graphs with sparse linkage information. We conducted experiments on three real-world network datasets, and the results indicated that the proposed method outperformed all benchmark approaches. language of the presentation: Japanese | |||
| AZUAJE SUAREZ GAMAR IVAN | D, 中間発表 | ソーシャル・コンピューティング | 荒牧 英治, 清川 清, 若宮 翔子, 矢田 竣太郎, She Wan Jou |
|
title: Cross-Modal Recommendation: Images for Expressive Writing and Music Discovery
abstract: Cross-modal recommendation refers to using information from one modality to enhance user experiences in another modality. In this study, we examine the potential of images to enhance two distinct creative processes: expressive writing and music discovery. The first examines the usage of synthesized images to enhance expressive writing and explores its effects on downregulating negative emotions during a fictional writing exercise. We developed a writing application that generates images in real-time based on user narratives. We examine the efficacy of the application in downregulating negative emotions and determine the user experience of the participants by qualitatively examining open-ended feedback. The second part explores the use of personal images for music discovery. We developed an application that matches the content of input images with song lyrics, enabling users to discover and immediately listen to relevant songs based on images of personal importance. We conducted semi-structured interviews to contrast the approach of exploring songs with images with current music discovery experiences. language of the presentation: English | |||
| 清基 英則 | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, 和田 隆広, 若宮 翔子, 矢田 竣太郎, 松田 裕貴 |
|
title: Estimating Tweet Directivity Using Linguistic Features
abstract: With the Covid-19 pandemic, governments and local authorities are often required to provide accurate and prompt information using social media. For the communication of such information, it is important for social media users themselves to know if they fall within the targeted demographics of these communications (age, gender, etc.), which we label here as ‘directivity’. Previous studies have mainly focused on examining the attributes of the information provider, but to our knowl- edge, there have not been any studies that examine the attributes of targeted users (receivers). In this study, we first assumed that tweets by magazine publishers are crafted for their targeted readership. We then collected tweets from the official Twitter accounts of these magazines and manually labeled the target age and gender of each magazine to create a dataset of tweet directivity. Using this dataset, we then classified the target user demographic (age group and gender) of these tweets through machine learning. For the additioanl experiment, we evaluated versatility of models through crowd-sourcing. We analyzed the results of this experiment and discussed the usefulness of our quantitative estimates of directivity. language of the presentation: Japanese 発表題目: 言語的特徴を用いたツイートの指向性推定 発表概要: 新型コロナウイルスの拡大に伴い，政府や自治体などはソーシャルメディアを用いた正確かつ迅速な情報発信を行うことが求められている.正確かつ迅速な情報発信のためには，特定の対象(年代や性別など)に向けて発信された情報を，その対象が自分に向けて発信されていると理解できるかどうか，すなわち「指向性」が重要である. 情報発信者の属性を特定する研究は多いが，情報が対象とする受信者の属性を特定する研究は見受けられない.本研究では，Twitter における雑誌の公式アカウントが発信するツイートは読者層向けに最適化されていると考え，雑誌の公式アカウントによるツイートを収集し，各雑誌の対象年齢と性別をラベル付けした指向性ツイートデータセットを構築した.このデータセットを用いて，対象ツイートがどの年齢，どの性別に向けられているものなのか機械学習モデルで分類した.追加検証では，クラウドソーシングを通して，モデルの汎用性の評価を行った．これらの実験結果を分析し，指向性の定量的測定がもたらす価値を考察した. | |||
| BASHIR HAMZA | 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, 向川 康博, 神原 誠之, 藤本 雄一郎, 澤邊 太志 |
|
title: Job status tracking for smart AR Task support system
abstract: This research presents an approach to automate augmented reality (AR) task support systems by integrating deep learning techniques for automated user activity tracking and task completion evaluation. The existing AR task support systems lack automation and user activity tracking. To address this issue, we used deep learning to detect the moment of completion of a sub-task and evaluate the task performed by the user. We collected a dataset specifically focused on the chair assembly task. By applying deep learning methodologies, we trained a neural network to recognize and understand human actions performed during the task. The deep learning model was trained to automatically detect the completion of each sub-task and evaluate the actions. language of the presentation: English | |||
| MUHAMAD ALDY BINTANG | 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, 安本 慶一, 神原 誠之, 藤本 雄一郎, 澤邊 太志 |
|
title: Correction for Improving Registration Position Accuracy of Virtual Objects in Indoor AR System abstract: With the support of Visual SLAM system, augmented reality (AR) system able to display digital information in 3D model onto the real environment using recent smartphone. However, SLAM model itself is not enough because the accumulation of camera pose error can lead to bigger misalignment of virtual objects, drawn in AR space. While the existing SLAM research focus on improving the camera system, we propose more simple yet applicable approach for existing AR-SLAM framework. Our method focuses on transform the virtual object into SLAM coordinate system correctly without correcting the distorted map or camera pose in SLAM. The method done by calculating the measurement error from marker recognition into weighted average solution for virtual object transformation. language of the presentation: English | |||