Yuyang Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yuyang Wang has a total of 170 co-authors. Here are the top co-authors by number of shared publications:
- Pan Hui
Hong Kong University of Science and Technology (Guangzhou)
15 shared publications - Jean-Rémy Chardonnet
Arts et Metiers Institute of Technology
10 shared publications - Danielle C. Maddix
AWS AI Labs ( 2795 Augustine Dr 95054 Santa Clara CA US
6 shared publications Hao Wang
AWS AI Labs AWS AI Labs Boson AI AWS AI Labs The Hong Kong University of Science and Technology
5 shared publications- Michael Bohlke-Schneider
Amazon
4 shared publications - Yitong Zhu
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China
4 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17963-17964
Momo's Digital Playground: Embodying Anonymous Identity and Collective Resonance through a 3D Interactive Installation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17963-17964
Open AccessMomo's Digital Playground: Embodying Anonymous Identity and Collective Resonance through a 3D Interactive Installation
- 12 August 2026
- Published by Taylor & Francis in International Journal of Human–Computer Interaction
- p. 1-35
From Adaptation to Intelligence: A Systematic Review of Data, Strategies, and Impact in Personalized VR
1. Virtual reality (VR) offers users an immersive experience by creating a virtual environment that leverages the combined power of hardware, software, tracking technology, and perceptual science (...
- 12 August 2026
- Published by Taylor & Francis in International Journal of Human–Computer Interaction
- p. 1-35
From Adaptation to Intelligence: A Systematic Review of Data, Strategies, and Impact in Personalized VR
1. Virtual reality (VR) offers users an immersive experience by creating a virtual environment that leverages the combined power of hardware, software, tracking technology, and perceptual science (...
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 13291-13292
Multi-Turn Reinforcement Learning for Large Language Models: From Theory to Practice with Amazon SageMaker AI
Production large language model (LLM) based systems such as coding agents, web navigators, and tool-calling assistants operate over multiple turns of interaction with users, tools, and environments. Pretrained LLMs, depending on their size, can either underperform in these settings due to misalignment with the system's interaction mechanics, or, when capable, incur prohibitive latency. Fine-tuning right-sized models addresses both accuracy and latency, but training such multi-turn agents requires Reinforcement Learning (RL), where the model acts as a policy optimizing long-horizon outcomes across sequential interactions. This poses challenges absent from single-turn settings: credit assignment over long trajectories, reward design for sparse and delayed feedback, state and context management as observation histories grow, environment scaling for parallel rollout collection, and training stability under prompt/environment distribution shift. This hands-on problem-solving tutorial provides both a rigorous algorithmic and practical introduction to multi-turn RL finetuning for LLMs. Using Amazon SageMaker AI, participants progress through four labs: (1) environment and reward function design, (2) multi-turn trajectory collection and Group Relative Policy Optimization (GRPO)-based training, (3) reward densification and credit assignment strategies, and (4) evaluation, failure diagnosis and deployment. We cover state-of-the-art multi-turn RL finetuning algorithms, turn-level vs. trajectory-level reward design, and production grade monitoring for reward hacking detection. The tutorial targets machine learning (ML) engineers, data scientists, and researchers who build agentic LLM systems. No prior RL experience is required. All materials will be publicly available on GitHub.
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 13291-13292
Open AccessMulti-Turn Reinforcement Learning for Large Language Models: From Theory to Practice with Amazon SageMaker AI
Production large language model (LLM) based systems such as coding agents, web navigators, and tool-calling assistants operate over multiple turns of interaction with users, tools, and environments. Pretrained LLMs, depending on their size, can either underperform in these settings due to misalignment with the system's interaction mechanics, or, when capable, incur prohibitive latency. Fine-tuning right-sized models addresses both accuracy and latency, but training such multi-turn agents requires Reinforcement Learning (RL), where the model acts as a policy optimizing long-horizon outcomes across sequential interactions. This poses challenges absent from single-turn settings: credit assignment over long trajectories, reward design for sparse and delayed feedback, state and context management as observation histories grow, environment scaling for parallel rollout collection, and training stability under prompt/environment distribution shift. This hands-on problem-solving tutorial provides both a rigorous algorithmic and practical introduction to multi-turn RL finetuning for LLMs. Using Amazon SageMaker AI, participants progress through four labs: (1) environment and reward function design, (2) multi-turn trajectory collection and Group Relative Policy Optimization (GRPO)-based training, (3) reward densification and credit assignment strategies, and (4) evaluation, failure diagnosis and deployment. We cover state-of-the-art multi-turn RL finetuning algorithms, turn-level vs. trajectory-level reward design, and production grade monitoring for reward hacking detection. The tutorial targets machine learning (ML) engineers, data scientists, and researchers who build agentic LLM systems. No prior RL experience is required. All materials will be publicly available on GitHub.
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 13285-13286
Taming Structured Data Foundation Models with AutoML: A Hands-On Guide
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 13285-13286
Open AccessTaming Structured Data Foundation Models with AutoML: A Hands-On Guide
- Open Access
- 13 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1553-1573
Dream the Dream: Futuring Communication between LGBTQ+ and Cisgender Groups in the Metaverse
- 13 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1553-1573
Open AccessDream the Dream: Futuring Communication between LGBTQ+ and Cisgender Groups in the Metaverse
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (8) , 7486-7499
Flow-Aware Diffusion for Real-Time VR Restoration: Mitigating Cybersickness With Enhanced Spatiotemporal Coherence
Cybersickness remains a critical barrier to the widespread adoption of Virtual Reality (VR), particularly in scenarios involving intense or artificial motion cues. Among the key contributors is excessive optical flow-perceived visual motion that, when unmatched by vestibular input, leads to sensory conflict and discomfort. While previous efforts have explored geometric or hardware-based mitigation strategies, such methods often rely on predefined scene structures, manual tuning, or intrusive equipment. In this work, we propose U-MAD, a lightweight, real-time, AI-based solution that suppresses perceptually disruptive optical flow directly at the image level. Unlike prior handcrafted approaches, this method learns to attenuate high-intensity motion patterns from rendered frames without requiring mesh-level editing or scene-specific adaptation. Designed as a plug-and-play module, U-MAD integrates seamlessly into existing VR pipelines and generalizes well to procedurally generated environments. The experiments show that U-MAD consistently reduces average optical flow and enhances temporal stability across diverse scenes. A user study further supports the finding that reducing visual motion defects can improve perceptual comfort and alleviate cybersickness symptoms. These findings demonstrate that perceptually guided modulation of optical flow provides an effective and scalable approach to creating more user-friendly immersive experiences. The code will be released at https://github.com/XXXXX (upon publication).
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (8) , 7486-7499
Flow-Aware Diffusion for Real-Time VR Restoration: Mitigating Cybersickness With Enhanced Spatiotemporal Coherence
Cybersickness remains a critical barrier to the widespread adoption of Virtual Reality (VR), particularly in scenarios involving intense or artificial motion cues. Among the key contributors is excessive optical flow-perceived visual motion that, when unmatched by vestibular input, leads to sensory conflict and discomfort. While previous efforts have explored geometric or hardware-based mitigation strategies, such methods often rely on predefined scene structures, manual tuning, or intrusive equipment. In this work, we propose U-MAD, a lightweight, real-time, AI-based solution that suppresses perceptually disruptive optical flow directly at the image level. Unlike prior handcrafted approaches, this method learns to attenuate high-intensity motion patterns from rendered frames without requiring mesh-level editing or scene-specific adaptation. Designed as a plug-and-play module, U-MAD integrates seamlessly into existing VR pipelines and generalizes well to procedurally generated environments. The experiments show that U-MAD consistently reduces average optical flow and enhances temporal stability across diverse scenes. A user study further supports the finding that reducing visual motion defects can improve perceptual comfort and alleviate cybersickness symptoms. These findings demonstrate that perceptually guided modulation of optical flow provides an effective and scalable approach to creating more user-friendly immersive experiences. The code will be released at https://github.com/XXXXX (upon publication).
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
GuideMe: A VLM-Based System Assisting Independent Smartphone Learning for Older Adults
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
Open AccessGuideMe: A VLM-Based System Assisting Independent Smartphone Learning for Older Adults
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (5) , 3411-3421
Aligning Gamification with Learner Motivation: Insights from VR-Based Learning Tasks
Virtual reality is increasingly adopted in education for its potential to create immersive, interactive learning experiences. Among various instructional approaches, gamification has emerged as a promising strategy to enhance learner engagement and motivation. While VR naturally provides a platform for game-based learning, current applications often overlook the underlying motivational mechanisms that influence learner behavior in these environments. Building on Self-Determination Theory, we examined how gamification and motivational framing influence VR learning. We designed a VR system featuring both gamified and non-gamified versions of two learning tasks—one culturally expressive (batik-based task) and one technically focused (code-based task)—to reflect different motivational framings. Our findings show that gamification selectively influenced intrinsic motivation components and played a dominant role in shaping learner satisfaction. A trend-level interaction suggested that combining extrinsic framing with gamification may increase satisfaction. Gamification improved overall experience and engagement, even though it elevated perceived workload. These results highlight the importance of aligning gamification strategies with learner motivation and underscore the need for more dynamic approaches to capture motivational processes in VR, offering design insights for more effective and psychologically attuned VR learning environments.
- 10 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Visualization and Computer Graphics
- Vol. 32 (5) , 3411-3421
Aligning Gamification with Learner Motivation: Insights from VR-Based Learning Tasks
Virtual reality is increasingly adopted in education for its potential to create immersive, interactive learning experiences. Among various instructional approaches, gamification has emerged as a promising strategy to enhance learner engagement and motivation. While VR naturally provides a platform for game-based learning, current applications often overlook the underlying motivational mechanisms that influence learner behavior in these environments. Building on Self-Determination Theory, we examined how gamification and motivational framing influence VR learning. We designed a VR system featuring both gamified and non-gamified versions of two learning tasks—one culturally expressive (batik-based task) and one technically focused (code-based task)—to reflect different motivational framings. Our findings show that gamification selectively influenced intrinsic motivation components and played a dominant role in shaping learner satisfaction. A trend-level interaction suggested that combining extrinsic framing with gamification may increase satisfaction. Gamification improved overall experience and engagement, even though it elevated perceived workload. These results highlight the importance of aligning gamification strategies with learner motivation and underscore the need for more dynamic approaches to capture motivational processes in VR, offering design insights for more effective and psychologically attuned VR learning environments.
- 6 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Navigation intelligente dans des environnements virtuels
La réalité virtuelle (VR) connaît un développement rapide grâce aux technologies de jeux vidéo 3D et de casques immersifs abordables, permettant d'interagir plus facilement avec des environnements virtuels. Au cours du processus d'interaction, l'utilisateur est au centre de l'application et, par conséquent, il est primordial de comprendre les facteurs humains affectant l'expérience et la satisfaction de l'utilisateur. Ces facteurs sont cruciaux pour concevoir une navigation virtuelle aussi naturelle que ce que les utilisateurs font généralement inconsciemment dans le monde réel.Cependant, les techniques de navigation dans les environnements virtuels peuvent être exigeantes sur le plan cognitif et induire un cybermalaise, conduisant au rejet des technologies de réalité virtuelle. Divers factors humains peuvent influencer l'expérience utilisateur, en particulier le niveau de cybermalaise pendant l'immersion en réalité virtuelle. Comme ces effets sont individuellement différents, son évaluation correcte est un préalable à la conception de différentes techniques de navigation. Pour utiliser ces différences individuelles et adapter la dynamique de navigation, nous avons introduit un modèle de logique floue pour analyser les caractéristiques humaines liées à l'apparition du cybermalaise et déterminer un indice personnalisé de cybermalaise. En plus des approches d'évaluation actuelles, nous avons démontré comment mesurer le cybermalaise et la charge cognitive en utilisant le deep learning (autoencodeur LSTM) et TOPSIS. La méthode d'évaluation proposée peut être considérée comme une amélioration des approches d'évaluation actuelles. Les résultats suggèrent que le deep learning représente une alternative innovante intéressante pour mesurer le cybermalaise en se basant sur des mesures comportementales comme la balance posturale, et que TOPSIS peut améliorer la précision de mesure de la charge cognitive.Les progrès réalisés pour accroître le potentiel de la réalité virtuelle reposent également sur notre capacité à développer des techniques de navigation qui peuvent atténuer le cybermalaise de manière efficace. Nous avons introduit la navigation adaptative dans le sens de la personnalisation et conçu quatre techniques différentes et originales de navigation tenant compte de différents facteurs humains. Dans un premier temps, nous avons proposé une navigation semi-automatique pour lisser les trajectoires de navigation lors des déplacements de l'utilisateur. Deuxièmement, connaissant l'importance des profils de vitesse de navigation sur l'expérience utilisateur, nous avons conçu un limiteur de vitesse capable de minimiser le jerk total lorsque l'utilisateur navigue dans des environnements virtuels. Troisièmement, le modèle de logique floue introduit pour déterminer un indice personnalisé de cybermalaise a été utilisé pour proposer une dynamique de navigation personnalisée. Enfin, sur la base de la théorie du contrôle PID et des réseaux de neurones, nous avons conçu une stratégie adaptative en ligne pour adapter la vitesse de navigation sur la base de l'état physiologique. Pour chaque technique de navigation, des expérimentations ont été réalisées pour valider leurs performances dans l'amélioration de l'expérience utilisateur, et les résultats montrent une réduction significative du niveau de cybermalaise.
- 6 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Navigation intelligente dans des environnements virtuels
La réalité virtuelle (VR) connaît un développement rapide grâce aux technologies de jeux vidéo 3D et de casques immersifs abordables, permettant d'interagir plus facilement avec des environnements virtuels. Au cours du processus d'interaction, l'utilisateur est au centre de l'application et, par conséquent, il est primordial de comprendre les facteurs humains affectant l'expérience et la satisfaction de l'utilisateur. Ces facteurs sont cruciaux pour concevoir une navigation virtuelle aussi naturelle que ce que les utilisateurs font généralement inconsciemment dans le monde réel.Cependant, les techniques de navigation dans les environnements virtuels peuvent être exigeantes sur le plan cognitif et induire un cybermalaise, conduisant au rejet des technologies de réalité virtuelle. Divers factors humains peuvent influencer l'expérience utilisateur, en particulier le niveau de cybermalaise pendant l'immersion en réalité virtuelle. Comme ces effets sont individuellement différents, son évaluation correcte est un préalable à la conception de différentes techniques de navigation. Pour utiliser ces différences individuelles et adapter la dynamique de navigation, nous avons introduit un modèle de logique floue pour analyser les caractéristiques humaines liées à l'apparition du cybermalaise et déterminer un indice personnalisé de cybermalaise. En plus des approches d'évaluation actuelles, nous avons démontré comment mesurer le cybermalaise et la charge cognitive en utilisant le deep learning (autoencodeur LSTM) et TOPSIS. La méthode d'évaluation proposée peut être considérée comme une amélioration des approches d'évaluation actuelles. Les résultats suggèrent que le deep learning représente une alternative innovante intéressante pour mesurer le cybermalaise en se basant sur des mesures comportementales comme la balance posturale, et que TOPSIS peut améliorer la précision de mesure de la charge cognitive.Les progrès réalisés pour accroître le potentiel de la réalité virtuelle reposent également sur notre capacité à développer des techniques de navigation qui peuvent atténuer le cybermalaise de manière efficace. Nous avons introduit la navigation adaptative dans le sens de la personnalisation et conçu quatre techniques différentes et originales de navigation tenant compte de différents facteurs humains. Dans un premier temps, nous avons proposé une navigation semi-automatique pour lisser les trajectoires de navigation lors des déplacements de l'utilisateur. Deuxièmement, connaissant l'importance des profils de vitesse de navigation sur l'expérience utilisateur, nous avons conçu un limiteur de vitesse capable de minimiser le jerk total lorsque l'utilisateur navigue dans des environnements virtuels. Troisièmement, le modèle de logique floue introduit pour déterminer un indice personnalisé de cybermalaise a été utilisé pour proposer une dynamique de navigation personnalisée. Enfin, sur la base de la théorie du contrôle PID et des réseaux de neurones, nous avons conçu une stratégie adaptative en ligne pour adapter la vitesse de navigation sur la base de l'état physiologique. Pour chaque technique de navigation, des expérimentations ont été réalisées pour valider leurs performances dans l'amélioration de l'expérience utilisateur, et les résultats montrent une réduction significative du niveau de cybermalaise.