Yiwen Yin
AI Agent Researcher | Ph.D. Candidate in Human-Computer Interaction at Tsinghua University
Education
Tsinghua University
2021 — PresentDirect Ph.D. Candidate, Human-Computer Interaction
Tsinghua University
2017 — 2021Bachelor's Degree, Computer Science and Technology
Experience
Pervasive Human-Computer Interaction Lab, Tsinghua UniversityPh.D. Researcher, AI Agents and Human-AI Interaction
2021 — Present- Research AI agents for task automation and collaboration, focusing on interactive knowledge learning, task modeling, UI agents, and Human-AI alignment.
- Build systems that move GUI automation beyond action replay toward interpretable cognitive task models, enabling agents to generalize user demonstrations to new goals and contexts.
- Study how proactive AI can support creative collaboration while preserving human agency, using empirical HCI methods and system-oriented research.
- Advised by Prof. Yuanchun Shi and Prof. Chun Yu.
Projects
TaskMind: From Operation to CognitionResearch System
2024 — 2025CHI 2025
- Proposed an LLM-enhanced GUI task automation system that identifies operation semantics and cognitive dependencies from user demonstrations.
- Represented tasks as user-interpretable graphs that can be edited for new goals and dynamically grounded to new parameters and interface contexts.
- Evaluated with 20 participants across predefined and customized tasks; TaskMind significantly outperformed an end-to-end LLM baseline in success rate and controllability.
TaskSense: Cognitive Chain Modeling and Difficulty EstimationResearch Framework
2024 — 2025- Designed Cognitive Chain, a framework that decomposes GUI task execution into cognitive steps such as finding, deciding, and computing.
- Developed an LLM-based pipeline to extract cognitive chains from task execution traces and estimate cognitive difficulty grounded in information theory.
- Showed that estimated cognitive difficulty correlates with user completion time and reveals capability gaps in state-of-the-art GUI agents.
Proactive AI as a Catalyst for Creativity?Human-AI Collaboration Study
2025 — 2026CHI 2026
- Investigated proactive AI support for collaborative story writing through a Wizard-of-Oz study comparing intrusive and non-intrusive suggestion styles.
- Found that proactive AI can accelerate writing and enhance creativity, while introducing a trade-off between AI involvement and perceived human agency.
- Derived design guidance for proactive suggestion timing and style to better balance creativity support with user control.
IMU Ring Interaction for Eyes-Free Smartphone OperationAccessibility Research
2019 — 2020IMWUT 2020
- Co-authored a ring-based input technique enabling visually impaired users to operate smartphones while keeping the phone in a pocket.
- Contributed to a participatory design and evaluation process for subtle one-handed gestures with audio feedback.
- The recognition model achieved 95.5% mean accuracy across 15 gesture classes, and the interaction was evaluated as efficient, private, and easy to use.
A CHI 2025 system that automatically models operation semantics and cognitive dependencies from user demonstrations, then executes interpretable task graphs with LLM support for robust GUI task automation.
A framework for estimating GUI task difficulty from cognitive chains rather than only motor actions, supporting user behavior analysis, GUI agent evaluation, and human-agent delegation.
A CHI 2026 study on proactive AI suggestion styles in creative collaboration, identifying design trade-offs between AI contribution, creativity, and perceived human agency.
Skills
Research Areas:
AI agents, GUI task automation, task and cognitive modeling, human-AI collaboration, Human-AI alignment, HCI user studies
Methods:
LLM-based system prototyping, Wizard-of-Oz studies, controlled user studies, task trace analysis, agent evaluation, interactive system design
Languages:
Chinese, English