| KPALEMON ABENA SAMUEL | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, | 松本 健一, | MD.Altaf-Ul-Amin, | 小野 直亮 | |
|
title: Enhanced Estimation of Bone Mineral Content from an X-ray Image Using Diffusion Models to Quantify Prediction Uncertainty for Osteoporosis Diagnosis.
abstract: Osteoporosis is the leading cause of bone fractures in the elderly, yet it often goes undiagnosed mainly due to the high cost and limited accessibility of dual-energy X-ray absorptiometry (DXA), the current gold standard for diagnosis. Additionally, deep learning algorithms like Generative Adversarial Networks (GANs) have shown promising results in diagnosing osteoporosis by estimating bone mineral content (BMC). However, previous GAN-based approaches have not addressed uncertainty estimation, which is critical to improving reliability in clinical decision-making. To build on previous work, we propose a novel method that combines random fast denoising diffusion probabilistic model (Random Fast-DDPM) for BMC estimation with a variance-based uncertainty quantification technique. Unlike GANs, which only estimate BMC, our method also captures prediction uncertainty, adding a layer of reliability to the diagnostic process. The proposed method achieved a high Pearson correlation coefficient, r=0.82 in BMC estimation and reported an overall uncertainty score of 0.189 with a weak positive correlation between BMC error and epistemic and aleatory uncertainty. Therefore further analysis is required in order to validate our uncertainty estimation *** language of the presentation: English | |||||||
| MUHAMMAD HENDRICK SEDAYU | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, | 松本 健一, | MD.Altaf-Ul-Amin, | 小野 直亮 | |
TitleExtraction of Molecules Feature for Classification Metabolic Pathway of Natural Compound AbstractNatural compounds are a promising source for new drug discovery, but their complex chemical structures make it challenging to understand how they are processed in the body. One key aspect is identifying the metabolic pathway each compound follows, which is essential for studying their biological roles and therapeutic potential. In this study, we propose a deep learning approach to automatically classify natural compounds based on their biosynthetic pathways using their molecular structures. We introduce a model called Augmented-Attention Graph Convolutional Network (AA-GCN), which combines graph convolutional layers to capture local structural features and a node-level attention mechanism to focus on the most relevant atoms in the molecule. This is further enhanced with global chemical descriptors, such as MACCS fingerprints and physicochemical properties. The resulting molecular representation combines both graph-based and domain-specific features, allowing the model to produce accurate and generalizable predictions. Our experiments show that AA-GCN effectively improves classification performance, especially for diverse or previously unseen compounds. LanguageEnglish | |||||||
| 山﨑 和真 | M, 2回目発表 | ソフトウェア工学 | 松本 健一, | 笠原 正治, | Raula Gaikovina Kula, | 嶋利 一真 | |
|
title: A Large-Scale Evaluation of a Merkle Tree-Based Method for Detecting Backward-Incompatible Changes in Libraries
abstract: The use of libraries is indispensable in software development. However, when a library is updated, unintended breaking changes may occur, potentially violating backward compatibility. If software that depends on such a library updates its dependencies, it may no longer function correctly. Prior studies have proposed a method for detecting backward-incompatible changes by capturing execution traces before and after a library update and comparing the Merkle trees constructed from those traces. However, only three evaluations using datasets have been conducted in previous research, and the datasets used did not include real-world cases where dependency updates in open-source software (OSS) actually caused errors. As a result, the effectiveness of the method has not been sufficiently demonstrated. To address this issue, our study conducts a large-scale evaluation using a new dataset. This dataset collects real-world cases from OSS projects in which dependency updates led to build or runtime errors, allowing for more realistic validation. Furthermore, in the process of conducting the large-scale evaluation, we automate the sequence of steps in the existing method to improve its applicability. language of the presentation: Japanese 発表題目: マークル木を用いたライブラリの後方非互換性特定手法に関する大規模有用性性評価 発表概要: ソフトウェアの開発にはライブラリの利用が欠かせない．ライブラリ自身がアップデートされた際に開発者の意図しない破壊的な変更が行われ，後方互換性が保たれていないことがある．後方互換性がないライブラリを使用しているソフトウェアがライブラリの依存関係を更新すると，ソフトウェアが正常に動作しなくなる可能性がある．先行研究ではアップデート前後の実行トレースを取得し，実行トレースから構築したマークル木の差分比較をすることでライブラリの非互換性を生じさせているメソッドを検知する手法が提案されている．しかし，既存研究においてデータセットを用いた検証は3件しか行われておらず，また用いたデータセットもOSS（Open Source Software）で実際に依存関係を更新したときにエラーが発生した事例ではないため，手法の有用性を十分に示すことができていない．そこで本研究では新たなデータセットを用いて大規模に検証を行う．本研究で用いるデータセットは，実際のOSSでライブラリの依存関係更新を行った際にエラーが発生した事例を収取しており，より現実に即した検証が可能である．また本研究では，大規模に検証を行う過程で既存手法の一連の手順を自動化する． | |||||||
| 金谷 温貴 | M, 2回目発表 | ディペンダブルシステム学 | 井上 美智子, | 笠原 正治, | WANG Wenyuan, | 江口 僚太, | 笹田 大翔 |
|
title: Almost Time-Optimal Loosely-Stabilizing Leader Election on Anonymous Arbitrary Graphs in the Population Protocol Model
abstract: In recent years, distributed systems where low-performance devices collaborate to perform a single task have attracted attention. Since low-performance devices have limited available resources, it is necessary to develop dedicated protocols and algorithms. The population protocol model[Angluin et al. 2004] represents a model for a network of anonymous low-performance devices. In this model, multiple mobile sensors, called agents, communicate with each other and progress the computation by updating their states according to predetermined transition rules (called protocols). It is known that when there is a unique leader in the population, various problems can be solved efficiently, which is why the leader election problem has been widely studied. Additionally, as the resiliency against transient faults, self-stabilization has been widely studied. However, in the population protocol model, it is known that the self-stabilizing leader election problem cannot be solved without the knowledge of the exact number of agents. Therefore, we consider loosely-stabilizing leader election[Sudo et al. 2009], which relaxes the closure property of self-stabilization in the population protocol model. Loose-stabilization means that the execution converges to the desired configuration within a relatively short time and maintains that configuration for a relatively long time. In this study, we propose a protocol with a convergence time close to the lower bound $\Omega(mN)$ shown by [Sudo et al. 2016] with increasing memory usage. This protocol uses $O(\Delta+\log{N})$ bits of memory, converges to a configuration with a unique leader in $O(mN\log N)$ expected steps, and maintains the configuration for $\Omega(Ne^{2N})$ expected steps, where $N$ represents an upper bound on the number of agents, while $m$ and $\Delta$ denote the number of edges and an upper bound on the maximum degree of the communication graph, respectively. This result shows improved memory efficiency compared to the M1 Colloquium, changing from $O(\Delta\log N)$ to $O(\Delta + \log N)$. language of the presentation: Japanese 発表題目: 個体群プロトコルモデルにおけるほぼ時間最適な匿名任意グラフ上の緩安定リーダー選挙 発表概要: 近年、低性能デバイス群が協調して1つの計算を行う分散システムが注目を集めている。低性能デバイスは使用できるリソースに限りがあるため、専用のプロトコルやアルゴリズムの開発が必要である。個体群プロトコルモデル[Angluinら 2004]は、匿名の低性能デバイス群のモデルである。このモデルでは、個体と呼ばれる複数のモバイルセンサーが相互に通信し、あらかじめ決められた状態遷移規則（プロトコル）に従って状態を更新し、計算を進める。 個体群にリーダー個体が1つだけ存在する場合、さまざまな問題が高速に解けることが知られているため、リーダー選挙問題は広く研究されている。また、一時故障に対する耐性として、自己安定性も広く研究されている。 ただし、個体群プロトコルモデルでは、正確な個体数を知らない限り、自己安定リーダー選挙問題を解くことができないことが知られている。そのため、自己安定性の要件の一つである閉包性を緩和した緩安定リーダー選挙[Sudoら 2009]を扱う。緩安定性を持つプロトコルは、比較的短い時間で目的の状態に収束し、その状態を比較的長い時間維持する。 本研究では、メモリの使用量を増やすことで、Sudoらが示した下界$\Omega(mN)$に近い収束時間を持つプロトコルを提案する。このプロトコルは、$O(\Delta+\log{N})$ビットのメモリを使用し、$O(mN\log N)$ステップでリーダーが1つの状態に収束し、その後$\Omega(Ne^{2N})$ステップにわたってその状態を維持する。 ここで、$N, m, \Delta$ は、それぞれ個体数の上界、通信グラフにおける辺数、及び最大次数の上界を指す。 この結果は、M1のColloquiumのときと比較してメモリ効率が向上しており、$O(\Delta\log N)$ から $O(\Delta+\log N)$ に改善されている。 | |||||||
| RAHMAN MD MUSTAFIZUR | 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, | 清川 清, | 澤邊 太志, | 山本 豪志朗, | Isidro Butaslac |
|
title: Experience Augmentation in Physical Therapy by Simulating Patient-Specific Walking Motions
abstract: In physical therapy, understanding and analyzing patient movements, especially impaired gait patterns, is crucial for effective rehabilitation. Traditionally, trainee therapists acquire these skills through hands-on experience with real patients and textbooks. However, these methods are limited by the availability of patients and the variability of impaired motions that therapists can observe. To address these limitations, we propose a novel system that allows therapists to learn from a wide range of impaired gait motions without being restricted by time, place, or patient availability. This system utilizes the HumanML3D dataset and a two-step framework combining text2length sampling and text2motion generation. In the first step, a classification model predicts motion length based on the input textual descriptions. For the second step, we use a temporal variational autoencoder (VAE) for generating varied and consistent 3D motion sequences. A key component of our approach is the utilization of residual vector quantization (RVQ) from the MoMask framework, which minimizes errors and enhances the precision of motion generation. Furthermore, a Masked Transformer ensures that the synthesized motion tokens are temporally consistent and contextually accurate. Our system, validated through the HumanML3D dataset, provides an immersive and interactive tool for physical therapists, enabling dynamic, patient-specific motion simulations in mixed reality environments. By bridging the gap between conventional methods and MR-assisted training, this approach uses interactive 3D representations to transform how therapists learn. It aims to revolutionize therapeutic training, making rehabilitation strategies more effective and personalized. language of the presentation: English | |||||||
| TUWAEMUESA THAPAKORN | 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, | 清川 清, | 澤邊 太志, | Isidro Butaslac, | 藤本 雄一郎 |
|
title:Evaluation of a VR Gaze Behavior Training System with LLM Conversational Agents Considering Conversational Context
abstract:Gaze plays a crucial role in effective communication, conveying attention, emotion, and intent. Previous research has identified common gaze patterns that facilitate smooth interpersonal interactions. In recent years, systems utilizing Virtual Reality (VR) technology have been proposed to help dialogue agents learn appropriate gaze behavior during conversations in virtual spaces. However, conventional systems often focus on rudimentary skills, such as agents engaging in fixed dialogues and contents, consistently directing their gaze towards the conversation partner's face. These systems have not been designed to acquire skills applicable to highly flexible situations like real-world conversations. Therefore, this research aims to construct a VR system that enables users to learn appropriate gaze behavior based on dialogue context and dynamic gaze guidelines, utilizing agents capable of flexible dialogue through recent Large Language Models (LLMs). The study will also investigate the effectiveness of this system. language of the presentation:English | |||||||
| LIU ZHONGXUE | 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, | 内山 英昭, | Perusquia Hernandez Monica, | 平尾 悠太朗 |
|
title:
abstract: I will present my experiment plan to evaluate an interface in Virtual Reality. I am currently running the experiment, if you want to participate, please skip my colloquium presentation and register for the experiment in this link: https://naist.sona-systems.com/default.aspx?p_return_experiment_id=41 language of the presentation: English | |||||||
| 金崎 知華 | 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, | 和田 隆広, | 澤邊 太志, | Isidro Butaslac | |
|
title: Designing Robot Behaviors with Consideration for Human Privacy and
Evaluating User Impressions
abstract: This study aims to design privacy-conscious robot behaviors and evaluate the impressions they create. In recent years, robots interacting with people in both domestic and public spaces have been increasing, resulting in more opportunities for them to encounter users’ confidential information. For example, the household robot Pepper can respond to health and emotional consultations, while the serving robot BellaBot interacts with customers in public spaces. In such situations, the information handled by robots may include privacy-related details that individuals do not want others to know. However, current discussions on privacy protection are heavily focused on technical measures such as encryption, data deletion, and control over data transmission, with insufficient attention paid to the behavioral aspects of robots, especially nonverbal behaviors. Therefore, this research investigates what kinds of behaviors robots should adopt in situations that require privacy consideration, how they should adjust their behaviors depending on the context, and how such behaviors affect user impressions such as trust, sense of security, and likability. Future plans include conducting a preliminary experiment observing how people behave when sharing secrets, an impression evaluation experiment using videos under four conditions related to nonverbal behaviors, and progressively carrying out human-robot interaction experiments to gain further insights. language of the presentation: Japanese 発表題目: プライバシーに配慮したロボットの行動設計と印象評価 発表概要: この研究は、プライバシーに配慮したロボットの行動設計と、その印象評価を目的とする。近年、家庭内や公共空間で人と関わるロボットが増加し、ユーザの秘密情報に触れる機会が多くなってきた。例えば、家庭用ロボットPepperは体調や感情の相談に応じることができ、配膳ロボットBellaBotは公共空間で顧客と接する。このような状況で、ロボットが扱う情報には個人が他者に知られたくないプライバシー情報が含まれる可能性がある。しかし、現状のプライバシー保護の議論は暗号化や情報削除、データ送信の制御など技術的処理に偏っており、ロボットの振る舞い方、特に非言語的な側面に関しては十分に議論されていない。そこで本研究では、プライバシー配慮が求められる状況でロボットがどのような振る舞いを取るべきか、状況に応じてどのように振る舞いを調整すべきか、そしてそれがユーザの信頼感や安心感、好感度といった印象にどのような影響を与えるかを調査する。今後は、秘密を話すときの人の振る舞いを観察する予備実験、非言語的振る舞いに対する印象評価を行う4条件の動画実験、さらにはロボットと人の対話実験を段階的に実施し、知見を得ていく予定である。 | |||||||