About Me
Hi, I’m Seyeon Choi, a PhD student at Seoul National University, affiliated with the Vision & Learning Lab. My research focuses on building generalizable embodied agents, with particular interest in grounding, interaction, and decision-making in complex environments.
I am currently exploring how large language models can move beyond static reasoning to act robustly as agents, especially under distribution shifts and unseen environments. My long-term goal is to develop agents that integrate symbolic reasoning with reinforcement learning to enable flexible, human-like behavior across diverse settings.
I have research experience across vision, language, and agent systems through internships and projects at multiple labs, and I enjoy translating theoretical ideas into practical, scalable systems.
Publications
ICML AIWILD 2026
WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents
Imene Kerboua, Fatemeh Pesaran Zadeh, Xing Han Lù, Weijian Qi, Alexander Miller, Junyi Song, Yunjia Tian, Dongjin Kang, Seyeon Choi, Marzia Nouri, Ewen Gueguen, Matteo Boglioni, Fengyuan Liu, Zeyi Liao, Mengqi Yuan, Yue Li, Alexandre Lacoste, Alexandre Drouin, Spandana Gella, Huan Sun, Gunhee Kim, Siva Reddy
[PDF] [Code (coming soon)]
ICML 2026
NAACL SRW 2024
Improving Multi-lingual Alignment Through Soft Contrastive Learning
Minsu Park*, Seyeon Choi*, Chanyeol Choi, Jun-Seong Kim, Jy-yong Sohn
* Equal contribution
Projects
Generalizable Web Agent (Capstone Design II)
Improving generalization of web agents to unseen websites via exemplar-based memory.
Designed a web agent that improves performance on unseen websites by constructing an exemplar memory during an exploration phase and leveraging it at inference time. This project focuses on robustness and cross-website generalization beyond in-domain training. Currently in progress as part of Software Capstone Design (II).
Robust Face Forgery Detection
Domain-robust deepfake detection using contrastive learning.
Built a robust deepfake detection system using supervised contrastive loss and domain-adversarial networks. This work was published and accepted at IPIU 2024, conducted as part of Software Capstone Design (I).
Multi-GPU Catmull–Clark Subdivision
Implemented the Catmull–Clark subdivision surface algorithm on multiple GPUs as part of CS267: Application of Parallel Computers. Focused on parallelization efficiency and scalability.
Experience
Vision & Learning Lab, Seoul National University
Research Intern (Advised by Gunhee Kim)
July 2024 – Feb 2025
Conducted research on web agents with a focus on grounding and interaction. Worked on agent generalization and robustness in real-world web environments.
Data and Language Intelligence Lab
Research Intern (Advised by Jinyeong Yeo)
June 2023 – June 2024
Supported experiments for CAFFEINE and Coffee, focusing on improving persona consistency in language models under temporal information changes. Participated in discussions on maintaining long-term persona quality.
Education
Seoul National University
PhD Student, IPAI (Vision & Learning Lab)
March 2025 – Present
PhD student advised by Prof. Gunhee Kim. Research focuses on embodied agents, grounding, and interaction, with a long-term goal of building generalizable decision-making agents.
Yonsei University
B.S. in Computer Science (Minor in Industrial Engineering)
March 2020 – February 2025
GPA: 4.08 / 4.3. Received multiple academic honors across semesters. Completed coursework and projects in machine learning, systems, and applied AI.
University of California, Berkeley
Exchange Student
January 2023 – May 2023
GPA: 3.83 / 4.0. Coursework included Abstract Algebra, Computer Security, Introduction to Analysis, and Application of Parallel Computing.
A Little More About Me
Outside of my research, I enjoy activities that help me recharge and stay curious.
- Cooking, especially Asian cuisine
- Exploring new restaurants and cafés ☕️
- Collecting and enjoying fragrances, with a preference for citrus 🍋🟩, floral 🌸, and soft musk notes
- Reading books in my spare time 📚
Below is a photo I took during one of my walks.
