ZhipengLi

PhD student at ETH

[ZhipengLi @ZhipengLiAsh]

The talk is on this Friday! Paper is available on arxiv: https://t.co/0D6cSyoVMH

“Preference-Guided Prompt Optimization for Text-to-Image Generation”
Zhipeng Li, Yi-Chi Liao, Christian Holz
ACM CHI 2026

[ZhipengLi @ZhipengLiAsh]

We introduce APPO that enables users to guide image generation using only their preferences, without writing or refining prompts.

With APPO, users simply need to (1) view generated results, (2) select the ones they like, and (3) receive improved results in the next iteration.

[ZhipengLi @ZhipengLiAsh]

APPO is not limited to image generation but can also be applied to a wide range of prompt-based generation tasks. We believe this is a step toward generative systems that adapt to users, not through better instructions, but by learning from their preferences.

[ZhipengLi @ZhipengLiAsh]

“Automating UI Optimization through Multi-Agentic Reasoning”
Zhipeng Li, Christoph Gebhardt, Yi-Chi Liao, Christian Holz
ACM CHI 2026

[ZhipengLi @ZhipengLiAsh]

AutoOptimization autoamtes UI adaptation through multi-agent reasoning.

Instead of manually defining objectives or selecting from many candidate layouts, users can simply describe what they want:

“I want all widgets in my field of view and the map on the TV.”

[ZhipengLi @ZhipengLiAsh]

This shifts UI adaptation from human-driven workflows to model-assisted ones. This is a step toward interfaces that not only adapt, but reason about user intent and optimize themselves.

[ZhipengLi @ZhipengLiAsh]

Under the hood, a sequence of VLM-based agents acts as a proxy designer: interpreting user preferences, defining objectives and parameters dynamically, and validating optimization results against user intent

[ZhipengLi @ZhipengLiAsh]

From this, the system:

1. detects ambiguity and asks clarification questions,
2. configures a personalized multi-objective optimization problem,
3. generates Pareto-optimal UI layouts,
4. and selects the solution that best matches the user’s intent.

[ZhipengLi @ZhipengLiAsh]

reference: Zhipeng Li, Yi-Chi Liao(@yichiliao), and Christian Holz(@cholz). Efficient Visual Appearance Optimization by Learning from Prior Preferences. ACM UIST 2025.
@ACMUIST
paper: https://t.co/SgLt3TutIb
project page: https://t.co/6sFdCkF9Q8

[ZhipengLi @ZhipengLiAsh]

📸 Are you adjusting photos? Forget rigid filters—just use Meta-PO!

Our #UIST2025 paper combines meta-learning and preference optimization, helping you discover personalized filters simply by selecting your favorite design 5–6 times.

👉 Read more: https://t.co/pUPy2a3LKP

[Yi-Chi Liao @yichiliao]

SIPLAB (all authors of my paper) @CSatETH at #CHI2025!

We are actively looking for Ph.D. Students to work on Computational HCI, e.g., Bayesian methods, optimization, sensing, and input decoding. Reach out to me or check https://t.co/xM8a6rGx48 for info!

[Yi-Chi Liao @yichiliao]

Come to my talk tomorrow at "Optimization with/for AI" (2:58 pm, G318).

Yi-Chi Liao, @paulstreli, @ZhipengLiAsh, Christoph Gebhardt, @cholz.
Continual Human-in-the-Loop Optimization

ACM CHI’ 25 @acm_chi
Paper: https://t.co/swXfdmu0ZW
Project: https://t.co/0cotOYw25z

[ZhipengLi @ZhipengLiAsh]

This work was done in collaboration with Yi Fei Cheng, @YukangYan and @da_lind.

[ZhipengLi @ZhipengLiAsh]

Have you ever been disrupted in VR interactions? We propose a new method to predict the noticeability of virtual animations by capturing the shift in visual attention. We believe our method can help content creators design just noticeable notifications in VR! #CHI2024

[ZhipengLi @ZhipengLiAsh]

The talk will be at 4:15 PM, Monday, at Room 311, in the "Understanding Immersive Experiences" session.

[Yu (Yuvia) Jiang @gottadive_yu]

Sharing our #chi2023 work HandAvatar! We enable precise, flexible, and embodied manipulation of non-humanoid avatars by mapping their complex structures to our highly dexterous and coordinated hands. w/ @ZhipengLiAsh, Mufei He, @da_lind, @YukangYan

[Arsenal @Arsenal]

So... How's your Saturday? 😉

#AFCvTHFC

[German Football @DFB_Team_EN]

It's that time. His final footsteps in 🇩🇪 colours. THANK YOU, LUKAS PODOLSKI! #PoldiButGoldie #GERENG 1-0