| 稲江 航輝 | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, | Sakriani Sakti, | 内山 英昭, | Perusquia Hernandez Monica, | 平尾 悠太朗 |
|
title: Emotional Expression of Avatars via Animal Ear Movemets Based on User Physiological Indices
abstract: In recent years, the adoption of VR has expanded rapidly. Within virtual spaces, sharing the same environment with conversation partners enables communication with people in distant locations as if they were face-to-face. Consequently, the importance of facilitating smooth communication through avatars in virtual spaces has grown. Emotional expression is crucial for smooth communication, and we primarily perceive it through facial expressions. Studies show that facial expressions also influence the recipient's impression when using avatars. In this study proposes a method that uses an avatar with animal ears on its head. It estimates the user's emotions from physiological indicators and expresses emotions by moving the animal ears according to the estimated emotion. By moving the animal ears in addition to the avatar's facial expressions, emotions are conveyed more effectively, leading to smoother communication. Furthermore, by not using key inputs for the animal ear movements, we believe it is possible to enhance emotional expression without burdening the user with facial expression manipulation. language of the presentation: Japanese 発表題目: ユーザの生理指標に基づいた獣耳運動によるアバター感情表現拡張 発表概要: 近年，VR の普及が急速に拡大している．仮想空間内では話し相手と空間を共有することで，離れた場所にいる人物であってもまるで対面しているかのようにコミュニケーションを行うことが可能である．これに伴い，仮想空間内でアバターを介して円滑なコミュニケーションを行うことの重要性が増している．円滑なコミュニケーションには感情表現が重要であり，我々は主に表情からそれを感じることができる．アバターを利用している場合にも，表情が受け手の印象に影響を与えることが示されている．本研究では，頭部に獣耳を持つアバターを利用し，ユーザの生理指標からユーザの感情を推定し，推定した感情に応じて獣耳を運動させることで感情表現を行う手法を提案する． | |||||||
| SUN RONGJUN | M, 2回目発表 | サイバネティクス・リアリティ工学 | 清川 清, | Sakriani Sakti, | 内山 英昭, | Perusquia Hernandez Monica, | 平尾 悠太朗 |
|
title: Editable Layer Generation for Animation Use of Character Illustrations
abstract: This research aims to convert character illustrations into editable layered structures that can be easily used for animation production. In a single illustration, elements such as hair, face, clothing, and accessories are usually integrated into one image, making it difficult to edit individual parts or directly use the illustration for animation systems such as Live2D. To address this problem, this research investigates a method for extracting semantic parts from an input character illustration and generating multiple transparent layers for complex regions such as hair and clothing. In particular, this study focuses on the fact that the required number of layers differs depending on the body part, and estimates an appropriate layer count using semantic information, shape features, and pseudo-depth information. By reducing missing structures and unnecessary empty layers, this research aims to generate layered representations that are easier to reuse for editing and animation production. language of the presentation: Japanese 発表題目: キャラクターイラストのアニメーション利用に向けた編集可能なレイヤー生成 発表概要: 本研究では、キャラクターイラストをアニメーション制作に利用しやすい編集可能なレイヤー構造へ変換することを目的とする。一枚絵のイラストでは、髪、顔、服、装飾品などが統合された状態で描かれているため、パーツ単位での編集や Live2D のようなアニメーション制作にそのまま利用することは難しい。そこで本研究では、入力されたキャラクターイラストから意味的なパーツを抽出し、さらに髪や服などの複雑な領域を複数の透明レイヤーとして生成する手法を検討する。特に、部位ごとに必要なレイヤー数が異なる点に着目し、意味情報、形状特徴、疑似的な奥行き情報を用いて適切なレイヤー数を推定する。これにより、必要な構造の不足や不要な空レイヤーの増加を抑え、編集やアニメーション制作に再利用しやすいレイヤー表現の生成を目指す。 | |||||||
| CUI ENCHENG | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, | Sakriani Sakti, | 若宮 翔子, | PENG SHAOWEN, | 久田 祥平 |
|
title: Single-Agent Generation Surpasses Multi-Agent Systems in Semantic Diversity abstract: Multi-Agent Systems (MAS) are commonlyused to improve reasoning diversity and robustness by simulating interactions among agentswith distinct roles. However, prior work oftenentangles the contribution of the multi-agentarchitecture with that of prompt conditioning,making the source of observed diversity gainsunclear. We address this confound with a controlled study on divergent thinking tasks, usingidentical prompt conditioning for MAS and single agent baseline. Under these matched conditions, single agent setups consistently outperform multi-agent systems in semantic diversity.We attribute this gap to information visibility:parallel agents often converge on overlappingideas, whereas a single agent model can condition on its own generation to avoid redundancy. We further find that a Multi-Output strategy, which prompts a single agent to producemultiple responses within a single inferencepass, achieves the highest diversity without degrading logical validity. Together, these resultspoint to a more efficient and effective way to expand diversity, with implications for the designof more efficient agentic frameworks. language of the presentation: English | |||||||
| XU JINSHA | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, | Sakriani Sakti, | 若宮 翔子, | PENG SHAOWEN | |
|
title: Confidence or Stubbornness? A Task-Space View of Multi-Agent Debate in LLMs abstract:Multi-agent debate (MAD) refers to a collaborative reasoning framework in which multiple LLM agents exchange candidate answers and revise their responses through discussion. While previous studies have shown that MAD can improve reasoning performance in some tasks, it remains unclear when debate supports confident convergence and when it instead leads to fixation on previously proposed answers. In this study, we investigate candidate-answer fixation in MAD from a task-space perspective, focusing on how answer-space structures affect agents’ belief revision. We first evaluate a MAS-based debate framework on the 20 Questions task, where agents must continuously update their hypotheses based on yes/no information, and compare it with a single-agent setting. Furthermore, we extend our analysis to mathematical reasoning, commonsense reasoning, and truthful question-answering tasks to examine whether this phenomenon depends on task characteristics. Our findings provide insights into the limitations of multi-agent collaboration and the conditions under which debate helps or hinders flexible reasoning. language of the presentation: English | |||||||
| 御前 賢斗 | 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, | 和田 隆広, | 澤邊 太志, | Isidro Butaslac |
|
title: Estimation of Passenger Stress Factors and Levels Based on the External Environment of Autonomous Driving
abstract: While autonomous driving technology currently prioritizes safety and efficiency, it often overlooks the psychological comfort of passengers, who can experience stress from interactions such as close encounters with other traffic participants. Previous studies have failed to directly link passenger stress with the dynamic external environment outside the vehicle. This research aims to estimate passenger stress factors and their levels during Level 4 and 5 autonomous driving by combining in-vehicle eye-tracking data with external front-view videos. To achieve this, the study creates a novel dataset by having participants watch driving scenes in a VR environment while measuring their electrodermal activity (EDA) to define stress and generate a "Stress Attention Map," ultimately contributing to the design of more comfortable routes for autonomous vehicles. language of the presentation: Japanese | ||||||
| 山田 黎也 | 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, | 和田 隆広, | 澤邊 太志, | Isidro Butaslac |
|
title:Understanding Peripheral Visual Environments through Visual Motion Cue Presentation for Motion Sickness Mitigation in Autonomous Vehicles
abstract: When passengers engage in non-driving tasks in autonomous vehicles, a mismatch between sensed vehicle motion and visual input can induce car sickness. This study aims to provide design guidelines for mitigating this through visual motion cues presented in the peripheral visual field. Prior work suggests acceleration-based cues reduce distraction while still mitigating sickness, and partial-field presentation may reduce distraction more than full-field presentation, yet the cue presentation range itself has rarely been systematically varied. This study therefore plans an experiment surrounding participants with three displays, presenting a central computation task with peripheral cues, to examine how the presentation range affects perception of cue movement direction and distraction (RQ1), and how predictive cue presentation can be adjusted to balance sickness mitigation against distraction (RQ2). language of the presentation: japanese 発表題目: 自動運転環境下における動揺病軽減のためのビジュアルモーションキュー提示手法による周辺視野環境の理解調査 発表概要: 自動運転車内で非運転タスクを行う際、身体が感じる車両の動きと視覚情報の不一致により動揺病（car sickness）が生じやすい。本研究は、周辺視野にビジュアルモーションキューを提示することでこれを軽減する設計ガイドラインの提供を目指す。先行研究では加速度ベースのキューが注意散漫を抑えつつ酔いを軽減できることや、部分視野提示が全視野提示より注意散漫を抑えられる可能性が示されているが、キューの提示範囲を体系的に検証した例は少ない。そこで、3枚のディスプレイで被験者を囲み、中央の計算タスク周辺にキューを提示する実験を計画し、提示範囲の違いが移動方向知覚と注意散漫知覚に与える影響（RQ1）、および酔い軽減と注意散漫のトレードオフを踏まえた予測的なキュー提示の調整方法（RQ2）を検討する。 | ||||||
| 谷 篤弥 | M, 2回目発表 | サイバーレジリエンス構成学 | 門林 雄基, | 和田 隆広, | 妙中 雄三 | |
|
title: Challenges and Proposed Solutions for gPTP Synchronization Algorithms in In-Vehicle Networks
abstract: In recent years, Time-Sensitive Networking (TSN), which provides highly accurate time synchronization, has been increasingly adopted in automotive networks. In TSN, precise time synchronization among Electronic Control Units (ECUs) is achieved over Ethernet using the generalized Precision Time Protocol (gPTP) defined in the IEEE 802.1AS standard. However, because gPTP assumes symmetric propagation delays along network paths, it is vulnerable to asymmetric delay attacks, in which an attacker intentionally introduces delays in only one communication direction to induce time synchronization errors. This study experimentally evaluates the synchronization errors that occur when delay symmetry is disrupted and proposes a novel time synchronization algorithm to mitigate this vulnerability. language of the presentation: Japanese 発表題目: 車載ネットワークにおけるgPTP同期アルゴリズムの課題と解決策の提案 発表概要: 近年、車載システムの内部ネットワークでは、高精度な時刻同期を実現する Time-Sensitive Networking（TSN）の導入が進んでいる。TSNでは、IEEE 802.1AS規格に基づく generalized Precision Time Protocol（gPTP）を用いることで、Ethernet上でECU（Electronic Control Unit）間の高精度な時刻同期を実現している。しかし、gPTPは通信経路における遅延の対称性を前提としているため、一方向の通信にのみ意図的な遅延を挿入する非対称遅延攻撃に対して脆弱であり、時刻同期誤差が生じる可能性がある。本研究では、遅延の対称性が崩れた環境下で発生する時刻同期誤差を実験的に評価し、その脆弱性を緩和する新たな時刻同期アルゴリズムを提案する。 | ||||||
| 林 純子 | D, 中間発表 | ソーシャル・コンピューティング | 荒牧 英治, | 田中 沙織, | 若宮 翔子, | 久田 祥平 |
|
title: Mirror, mirror on the LLM: Evaluating LLM Responses to Harmful Body-Image Beliefs abstract: LLM-based chat systems are increasingly being used as sources of mental health-related support, but empathetic and supportive responses are not always clinically safe. In particular, harmful beliefs about body image should not be validated regardless of user context. Recent studies have shown that when LLMs are given information about users, they are more likely to generate responses that align with users’ thoughts and emotions. However, it remains unclear whether differences in user context lead to the validation of harmful beliefs in clinically sensitive domains. Therefore, this study evaluates how LLM responses to the same harmful body-image belief change depending on user context. Using user utterances generated based on SATAQ-4R, we set six context conditions: no personal information, neutral personal information, apparently safe body-related information, general vulnerability, body-image vulnerability, and explicit clinical-risk information. We collected responses from multiple LLMs and evaluated them in terms of body-image belief reinforcement, emotional validation, weight-loss/appearance advice, and risk sensitivity. The results showed that explicit clinical-risk information increased risk sensitivity, whereas apparently safe body-related context did not sufficiently increase caution and sometimes led to appearance-oriented advice or validation of harmful beliefs. These findings suggest that LLM safety evaluations should consider not only what users ask, but also the user context surrounding their utterances. language of the presentation: Japanese | ||||||
| 小泉 孝太朗 | M, 2回目発表 | 脳・行動モデリング（計算神経科学） | 田中 沙織☆, | 川鍋 一晃, | 杉本 徳和, | 荻島 大凱 |
|
title: Alignment between implicit and explicit metacognition promotes abstract learning
abstract: Humans can adapt to the unknown environments by learning from small size samples so efficiently. This remarkable ability is often attributed to abstraction: the capacity to extract only relevant information from complex sensory information. Previous research has shown that humans adaptively shift from comprehensive to abstract learning strategies with experience. While confidence is hypothesized to be relevant to this shift, the computational mechanisms by which confidence signals modulate learning strategies remain unknown. To elucidate this mechanism, we focused on confidence, which serves as the quantitative readout of metacognition, defined as the capacity to monitor and control one's own cognitive processes. We posited that this confidence signal acts as the critical input for regulating behavior. In this study, we investigated whether and how metacognition regulates the strategies of abstract learning. language of the presentation: English | ||||||
| 東 青空 | M, 2回目発表 | 脳・行動モデリング（計算神経科学） | 田中 沙織☆, | 川鍋 一晃, | 杉本 徳和, | 荻島 大凱 |
|
title: Input Optimization of Optogenetic Neural Stimulation via Reservoir Computing
abstract:Recent progress in the field of optogenetics has allowed the neuroscience researchers to manipulate specific neuronal populations with millisecond precision to investigate causal circuit mechanisms. Despite these advances, there still remains the inverse problem of an input optimization—determining the exact stimulation pattern that will drive a neural circuit into a desired target activity. Even in computational brain models where the entire neural wiring diagram is known, identifying the optimal input signals remains challenging. The difficulty lies in the complex non-linear dynamical relationships between input and output neurons. Here, we propose a reservoir computing system to solve this input optimization problem. Reservoir computing is a machine learning framework that excels at processing temporal data and modeling chaotic, non-linear dynamical systems with low training complexity. Uniquely, the reservoir’s high-dimensional recurrent connectivity allows it to act as a latent dynamical space, effectively learning the inverse transfer function of the biological network through computationally efficient readout training. We plan to experimentally validate this framework using computational brain model of a fruitfly called FlyWire. This framework should provide a robust theoretical basis for optogenetic control with a possible capability of closed-loop dynamical control. language of the presentation: English | ||||||
| 阿部 龍之介 | M, 2回目発表 | 数理情報学 | 池田 和司, | 田中 沙織, | 久保 孝富, | LI YUZHE |
|
title: From Axes to Anchors:
Finite-Anchor Geometry for Text Embeddings
abstract: Representation spaces in language models are often interpreted geometrically, and task-relevant information is often linearly recoverable. However, linear recoverabil- ity does not imply that such information is organized along raw coordinate axes. We therefore study a stronger and more relational diagnostic: whether task-label struc- ture survives when embeddings are forced to be represented as convex mixtures of a finite set of anchors. Instead of seeking global coordinate directions aligned with label information, we fit an unsupervised finite-anchor convex decomposition and analyze its anchor coefficients, convex reconstruction, and residuals. We find that task-label structure selectively survives this constraint: topic datasets with embedding-specialized representations often form compact anchor-relative regions or low-dimensional faces, whereas finer-grained or multilabel emotion settings show broader anchor support. Random-anchor controls indicate that this alignment is not an artifact of arbitrary convex mixtures. We therefore position finite-anchor convex decomposition as a diagnostic for locating task-label information under a strong geometric constraint. We further show that anchor mixtures preserve rela- tional label structure: coarse topics can be separable while remaining multi-facet unions of finer category-level regions. language of the presentation: English | ||||||