Date：2025/10/17 15:40-17:00
Location：R103, CSIE
Speakers：Dr. Yi-Chi Liao
Host：鄭龍磻教授
Abstract:
Designing interactive systems is, at its core, a challenging optimization problem. Whether calibrating AR/VR devices, configuring wearable sensors, or steering AI to generate content, design ultimately involves finding effective solutions within vast and complex design spaces. Traditional User-Centered Design workflows evaluate a limited set of candidates proposed by designers to approximate a “good design” before deployment. Not only does this risk omitting promising design candidates, but this “one-design-fits-all” approach also fails to scale to emerging interactions and AI-embedded systems, given the user diversity and the challenges of predefining all input-output behaviors. For example, many AR/VR interactions require personalization before use; in domains such as gaming or visual design, customization directly enhances performance. Moreover, AI-embedded systems are driven by natural-language instructions rather than pre-specified interactions, making traditional pre-design infeasible. In this talk, I argue that this is a pivotal moment to reimagine interface design. The future of Human-Computer Interaction (HCI) is not about crafting fixed, polished interfaces, but about creating systems that continuously redesign themselves for diverse users and contexts. To realize this vision, I develop Human-in-the-Loop (HITL) intelligent systems that learn from human behavior and preferences, and adapt dynamically during use. In such systems, interaction itself becomes an ongoing optimization process. I demonstrate how HITL systems can scale across diverse design tasks, ranging from fine-tuning personal inputs to preference-guided design optimization and steering generative models. The talk is organized around three pillars of my work: (1) Workflows that enable general-purpose HITL systems; (2) Efficient HITL methods powered by learned priors; (3) Generalizable synthetic users that further scale adaptation. Together, these pillars point toward a future of interactive systems that evolve with and for humans, in real time.
Biography:
Yi-Chi Liao is a postdoctoral fellow at SIPLAB, ETH Zürich, supported by the ETH Postdoctoral Fellowship. His research lies at the intersection of Human-Computer Interaction, machine learning, and design optimization. He develops human-in-the-loop systems that adapt, optimize, and co-create with humans. He also serves as Associate Chair for ACM CHI and ACM UIST. Previously, he was a postdoctoral fellow at Saarland University and a research intern at Meta Reality Labs. He received his Ph.D. from Aalto University. More information is available at: https://yichiliao.com