Hi, I’m Vivian (Yiwei) Chen, a second-year Ph.D. student in Computer Science at the University of California, San Diego. I am fortunate enough to be jointly advised by Prof. Taylor Berg-Kirkpatrick and Prof. Rose Yu. Before that, I received my B.A. in Computer Science, Mathematics, and Linguistics from Cornell University.
Research
My research centers on LLM agents and AI for science. I’m particularly interested in agentic evaluation — understanding how agents acquire evidence, reason causally, and commit to decisions under uncertainty. I’m also interested in reinforcement learning for agents, causal and statistical reasoning, data-efficient modeling, and low-resource NLP.
My current work at UCSD includes ACED-Bench (NeurIPS 2026), a benchmark for evaluating whether LLM agents can actively acquire causal evidence and act on it, and Zephyrus, an agentic framework for weather science — both advised by Prof. Taylor Berg-Kirkpatrick and Prof. Rose Yu. Previously, I worked with Prof. Kilian Weinberger on diffusion-guided language modeling (Findings, ACL 2024) and with Prof. Yoav Artzi on retrospective learning from interactions (ACL 2025 Oral).
News
- 09/2026: Our paper ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence was accepted to NeurIPS 2026. See you in Atlanta!
- 06/2026: I started a summer internship at Amazon as an Applied Science Intern, working on LLM agents.
- 04/2026: I presented a poster on ACED-Bench at Amazon Research Day in Palo Alto.
- 12/2025: Our paper Zephyrus: An Agentic Framework for Weather Science was presented at the NeurIPS 2025 Workshop on AI4Science.
- 06/2025: Our paper Retrospective Learning from Interactions has been selected for an oral presentation at ACL 2025.
- 05/2025: I graduated from Cornell with a B.A. in Computer Science, Math, and Linguistics.
Selected Publications
Industry
Built an LLM agent that explores historical transaction data, forms hypotheses about emerging patterns, and turns them into interpretable rules. Also designed the RL rewards and training setup, using a rule-evaluation simulator as the environment.
Awards
Service
- NeurIPS 2025 Workshop on AI for Science
- NeurIPS 2025 Workshop on Scaling Environments for Agents (SEA)
- CS 4782: Introduction to Deep Learning
- CS 4780: Machine Learning (twice)
- CS 4740: Natural Language Processing
- CS 2800: Discrete Structures
- CS 2802: Honors Discrete Structures
- INFO 1998: Introduction to Machine Learning (student-led course)
- Student Volunteer, ACL 2026
- Peer Mentor, Women in Computing at Cornell (WICC)
- Instructor, Girls Who Code
- Translator, Cornell Translator Interpreter Program (2023 James A. Perkins Prize recipient)
Miscellaneous
I'm a huge musical theatre fan. I worked for Tony Awards · China as a subtitle translator in 2024.
I like singing, reading, and watching movies. I quite enjoy learning languages: my Japanese is at an intermediate level, and I'm currently learning Spanish.
For sports, I like skiing and figure skating, and I play table tennis. I also started watching fútbol.
In my spare time, I also do a bit of seal cutting and crocheting, and I'm looking forward to learning knitting.
