主題演講

Professor Sabarish Babu

Clemson University, USA

In this keynote, I will first discuss lessons learned from a body of research on emotion contagion and visual attention in human-virtual human interaction. In a virtual human simulation designed to educate nurses in recognizing the signs and symptoms of rapid deterioration, we investigated the effects of animation, appearance and interaction fidelity on the emotional reactions and visual attention behaviors of trainees in simulated dyadic and crowd scenarios. Our results and lessons learned have implications to the design of virtual humans in inter-personal simulations for personal space education. Next, I will discuss a body of work investigating dimensional symmetry and interaction fidelity continuum on near-field fine motor skills training in VR for technical skills education in the aviation and automotive curriculum. We designed and evaluated the effects of interaction, dimensional and system fidelity in near-field virtual reality simulations for motor skills training and education, in domains such as precision metrology and mechanical skills acquisition. Finally, I will discuss our some of our key findings in static and dynamic affordances in VR and MR, as well as perception-action coordination research with implications for near and medium field psychomotor skills training and education. I will end the talk summarizing our contributions, and highlight the key takeaways and recommendations for the design of virtual human and mixed reality simulations for near-field fine motor skills training and inter-personal skills education.

講者簡歷

Professor Tzu-Mao Li(李子懋)

University of California San Diego, USA

While neural networks have become powerful tools for processing visual data, their generality raises several challenges. Firstly, most modern architectures work on 2D, and it is difficult to embed 3D knowledge. Secondly,  neural networks are by design over-parametrized and have millions or billions of parameters. It is challenging to make the networks run fast for high-resolution images and videos on mobile devices. Finally,  neural networks are difficult to debug and control as their behaviors are mostly governed by their parameters and the training data. On the other hand, classical visual computing algorithms that explicitly model the computation are less impacted by these issues. Still, they often do not apply as broadly as modern data-driven methods. A major focus of our research is to connect classical graphics algorithms with modern data-driven methods, by making graphics algorithms differentiable to enable optimization and inference. Making graphics algorithms differentiable lead to new challenges: How do we derive the correct derivatives in the first place, when there can be discontinuities and boundary conditions involved? How do we efficiently compute the derivatives? How do we build systems to make derivation and implementation of differentiation easier? I will talk about our recent efforts into addressing these challenges. These research efforts include contributions in fields of forward/inverse rendering, image processing, physical simulation, and programming languages/systems.

講者簡歷