| 宮井 菜名子 | M, 1回目発表 | 数理情報学 | 池田 和司 | 清川 清 | 久保 孝富 | 日永田 智絵 | Li Yuzhe |
|
title: Embedding method for dog behavior based on h/BehaveMAE
abstract: Dogs and humans have established a long-standing mutually beneficial relationship through coexisting over time, during which dogs have developed social cognition to understand human instructions and gaze. To investigate this social cognition, dog behavior analysis has been conducted. However, conventional studies required pre-labeled videos of behaviors, resulting in high annotation costs and the inability to analyze undefined behaviors. To address these challenges, this study proposes a method to automatically identify behaviors with similar features from videos. Pose trajectories estimated using DeepLabCut are input into h/BehaveMAE, which outputs frame-wise behavior embeddings, enabling the identification of behaviors directly from the videos. In the future, as a fundamental validation of h/BehaveMAE, training and quantitative evaluation will be conducted using datasets including Animal Kingdom. Subsequently, further validation will be performed using data collected independently. This research is expected to contribute to the advancement of dog studies and applications such as supporting the diagnosis of dog diseases. language of the presentation: Japanese 発表題目: h/BehaveMAEに基づくイヌの行動の埋め込み手法 発表概要: イヌとヒトは共に生活する中で互恵的な関係を築いており、その過程でイヌはヒトの指示や視線を理解する社会的認知能力を発展させてきた。このような社会的認知能力を調べるために、イヌの行動分析が行われてきた。しかし、従来の手法では、行動が事前にラベル付けされた動画を用意する必要があり、アノテーションのコストが高いこと、未定義の行動を分析することができないことなどが課題となっていた。本研究では、これらの課題を解決するために、撮影した動画からDeepLabCutを用いて推定された姿勢情報の軌跡を自己教師あり学習手法であるh/BehaveMAEに入力し、行動の埋め込みをフレームごとに出力することで、動画から自動的に似た特徴を持つ行動を検出する方法を提案する。今後、h/BehaveMAEの基礎検証として、Animal Kingdomを含むデータセットを用いたトレーニングと定量的な評価を実施し、その後、自身で収集したデータを用いてさらなる検証を行う予定である。本研究により、イヌの研究の発展や、イヌの疾患診断支援などへの応用を期待している。 | |||||||
| 宮川 翔太 | M, 1回目発表 | 数理情報学 | 池田 和司 | 作村 諭一 | 久保 孝富 | 日永田 智絵 | Li Yuzhe |
|
title: Defecation prediction system based on bowel sound measurement
abstract: With aging, defecation disorders such as constioation and fecal incontinence are increasing, leading to significant physical and mental burdens in elderly care. Developing a wearable device capable of predicting defecation timing is expected to alleviate caregivers' burdens and improve the quality of life for the elderly. Previous studies have suggested differences in bowel sounds before and after defecation. Building on these findings, this study aims to develop a system that utilizes bowel sound data to predict defecation timing in advance. language of the presentation: Japanese 発表題目: *** この部分を発表題目に *** 発表概要: *** この部分を発表概要に *** | |||||||
| CHU SHIWEN | M, 1回目発表 | 数理情報学 | 池田 和司☆ | 川鍋 一晃 | 杉本 徳和 | ||
|
title: Enhancing cross-subject emotion classification performance using resting-state EEG and multimodal features abstract: Emotion classification (EC) using EEG signals has long attracted researchers due to its potential practicality and inherent interest. However, inter-subject variability caused by individual differences remains a critical challenge for developing more effective classification methods. Inspired by advances in other areas of EEG research, we propose leveraging the commonly available resting-state EEG (RS EEG) signals to capture subject-specific characteristics, addressing the issue of inter-subject variability. This research introduces an emotion classification framework that separates features into EC_task-related and subject-related components and calibrates these using features extracted from RS EEG to enhance the model’s cross-subject generalization ability. Furthermore, acknowledging cultural differences in emotional expression, features extracted from synchronized facial videos are integrated to provide complementary information, improving emotion recognition accuracy. This research aims to contribute to the development of methods for analyzing emotions that take into account the various types of differences between subjects. language of the presentation: English | |||||||
| 山口 晴久 | M, 1回目発表 | 数理情報学 | 池田 和司☆ | 作村 諭一 | 川鍋 一晃 | 田中 沙織 | |
| 成田 大祐 | M, 1回目発表 | 計算行動神経科学 | 田中 沙織 | 清川 清 | 川島 一朔 | |
|
title: Detecting Mind Wandering With Minimal Equipment
abstract: Mind wandering occurs when a person's attention drifts away from the task at hand, leading to reduced focus and productivity. This study investigates the detection of mind wandering using minimal equipment, focusing on brain signals and eye movements. By utilizing compact devices, such as in-ear EEG sensors and eye-tracking technology, the research explores the feasibility of accurately identifying mind wandering in everyday scenarios. Results from testing 40 subjects are presented, highlighting the potential for practical applications of this technology in daily life. language of the presentation: English | ||||||
| 我妻 諒治 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元 | 和田 隆広 | 市川 昊平 | 柏 祐太郎 |
|
title: A Study on Quality Analysis of Automated Driving Software Repositories
abstract: Ensuring and improving software quality is essential for guaranteeing the safety and reliability of automated driving systems. In particular, it is crucial to develop a deep understanding of the quality of software that supports automated driving functions and performance. Previous research has primarily focused on quality analysis of foundation software such as Autoware and Apollo. However, when implementing these systems in actual vehicles, it is common practice to extend the foundation software based on vehicle-specific requirements. This research focuses on such extended software and aims to elucidate the quality characteristics of extended software through quality analysis of automated driving software repositories. language of the presentation: Japanese 発表題目: 自動運転ソフトウェアリポジトリの品質分析に関する研究 発表概要: 自動運転ソフトウェアにおいて安全性と信頼性を保証するためには，ソフトウェア品質の確保と向上が不可欠である．特に，自動運転の機能や性能を支えるソフトウェアの品質を深く理解することが求められる．これまでの研究では，AutowareやApolloなどの基盤ソフトウェアに対する品質分析が主に行われてきた．しかし，実際の車両に搭載される際には，車両固有の要件に基づき，基盤ソフトウェアが拡張されることが一般的である．本研究では，そのような拡張されたソフトウェアに注目し，自動運転ソフトウェアリポジトリを対象に品質分析を行うことで，拡張ソフトウェアの品質特性を解明することを目的とする． | ||||||
| 加藤 陸 | M, 1回目発表 | ソフトウェア設計学 | 飯田 元 | 林 優一 | 市川 昊平 | 柏 祐太郎 |
|
title:A Study on Vulnerability Prediction Using Continuous Fuzzing Data
abstract:In recent times,a considerable number of significant security issues resulting from vulnerabilities in open-source sofrware have been reported. Fuzzing has emerged as a prominent approach in software testing,offering a solution to the aforementioned challenges.Fuzzing is a testing method that is employed to identify vulnerabilities by deliberately inducing the exceptions in software by providing inputs that may potentially lead to errors.However,fuzzing is known to be a computationally expensive and time-consuming process,resulting in a significant delay between the addition of vulnerable code to a repository and the discovery of the associated vulnerability through fuzzing.In this study,we propose a method for expedient discovery of vulnerabilites as soon as they are added to the repository.Specifically,we construct a machine learning model that identifies vulnerabilities in program changes based on historical trends,while leveraging the execution results during the fuzzing process. language of the presentation:Japanese 発表題目:Continuous Fuzzing データを用いた脆弱性予測に関する研究 発表概要:近年，オープンソースソフトウェアの脆弱性が要因となる大規模なセキュリティに関する問題が多数報告されている.この問題を解決するテスト手法としてFuzzingが注目されている．Fuzzingとは問題を引き起こす可能性がある入力をソフトウェアに与え，意図的に例外を出力させることで，脆弱性を発見するテスト手法である．しかしながら，Fuzzig は膨大な計算リソースと実行時間を必要とすることで知られる．つまり，脆弱性のあるコードがリポジトリに追加されてから，Fuzzingが完了し，脆弱性が発見されるまで大きなラグが生じる．本研究では，脆弱性のあるコードがリポジトリに追加れた段階で発見する手法を提案する．具体的には，Fuzzing途中における実行結果を活用しながら，過去の傾向からプログラムの変更に含まれる脆弱性を発見する機械学習モデルを構築する． | ||||||