My name is Yixin (Elaine) Wan and I am a final-year PhD candidate in Computer Science at UCLA. I am fortunate to be advised by Professor Kai-Wei Chang and be a member of the UCLANLP research group. I completed my B.S. in Applied Mathematics (double major in Economics) also at UCLA. Go Bruins!

I build and improve multimodal LLMs through (a) mid-training and post-training, especially through large-scale synthetic data construction, (b) workflow / harness evolvement for LLM agents on challenging new tasks, and (c) interpretable and faithful evaluation on novel tasks. My research is driven by diagnosing and resolving new model failure modes on multimodal understanding, controllability, and trustworthiness.

I have joined FAIR @Meta Superintelligence Labs as a Research Scientist Intern in June 2026 in Menlo Park, CA, working on multi-task large-scale MLLM joint mid-training. Previously, I have interned at Tencent Hunyuan, Amazon AGI, and Microsoft Research Asia (MSRA).

I am actively looking for full-time opportunities starting EOY 2026 or 2027 spring. Feel free to reach out!

News

• [2026/09] 📢 🤔 New Work: How well can SOTA Multimodal LLM recognize fine-grained visual differences in two images? Introducing VDiff-Bench: Diagnosing and Improving Fine-Grained Visual Difference Understanding! Check it out: Paper link . Also see our Project Page!
• [2026/07] I served as a Program Chair for the second time at the ACL 2026 TrustNLP Workshop. Grateful for all the support!
• [2026/07] I gave a talk on Reflective & Bias-Aware Knowledge Control at the ACL 2026 Tutorial: Knowledge Control for Responsible Generative AI, where I highlight agentic mitigation and our related recent work as a promising research direction in trustworthy AI.
• [2026/06] I joined FAIR @Meta Superintelligence Labs as a Research Scientist Intern at Menlo Park, CA. I will be working on multi-task Multimodal LLM mid-training. Excited to explore the Bay!
• [2026/04] 2 papers accepted to ACL 2026 Main conference and 1 accepted to Findings 🎉! More details coming soon. See you in San Diego 😊
• [2026/02] MotionEdit is accepted to CVPR 2026 🎉! See you in Denver 😄
• [2025/12] 📢 ⭐️ New Work on Motion Image Editing with Tencent Hunyuan released! Check it out: MotionEdit: Benchmarking and Learning Motion-Centric Image Editing . Also check out more demos on our Project Page!
• [2025/09] I am humbled to be awarded the 2025 Amazon AI Fellowship 🎉!
• [2025/08] 3 papers on LLM Unlearning and Fairness accepted to EMNLP 2025 Findings 🎉! Check them out:
  • Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning
  • LUME: LLM Unlearning with Multitask Evaluations
  • Fact-or-Fair: A Checklist for Behavioral Testing of AI Models on Fairness-Related Queries
• [2025/06] Started my internship at Tencent Hunyuan (Bellevue)!
• [2025/05] 2 long papers on Fairness accepted to ACL 2025 Main 🎉! Check them out:
  • The Male CEO and the Female Assistant: Evaluation and Mitigation of Gender Biases in Text-To-Image Generation of Dual Subjects
  • White Men Lead, Black Women Help? Benchmarking and Mitigating Language Agency Social Biases in LLMs
• [2025/05] Arrived at Albuquerque! I am serving as the Program Chair at NAACL 2025’s Trustworthy NLP (TrustNLP) Workshop.
  • I am also presenting my recent work on LLM unlearning with Amazon: Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning