Yingjia Wan

(Other name: Alisa)

About

I am Yingjia Wan, a second-year PhD student at UCLA studying computational communication, advised by Elisa Kreiss in the Computation and Language for Society (Coalas) Lab and the UCLA NLP group.

My research interests include Large Language Model (LLM) reasoning, AI for math (specifically, autoformalization in Lean 4), multimodality, and reliable agentic systems. My skillset emphasizes post-training and evaluation methods, combining qualitative and quantitative analysis. Details in CV.

I am seeking a 2027 summer internship, and open for academic collaboration. Contact at email.

Education

University of California, Los Angeles (PhD), 2025-current
University of Cambridge (MPhil), 2022-2023
University of Macau (BA), 2017-2021

Featured Publications

Yingjia Wan, Lin Lin, Elisa Kreiss

September, 2026 COLM 2026

How Humans and LLMs Read Gender into "Gender-Neutral" Physical Descriptions

We introduce GAPA, a dataset of physical attributes paired with 14,706 human gender-association ratings, showing that seemingly “gender-neutral” descriptions still carry systematic gender associations that 16 LLMs only partially and unevenly capture. We train a proxy model to predict these associations at scale.

Haocheng Wang, Baiyu Huang, Yingjia Wan, Xiao Zhu, Xiaoyang Liu, Yinya Huang, Zhijiang Guo

July, 2026 ICML 2026 (& Official Challenge Benchmark for the 3rd AI4MATH Workshop @ICML2026)

FormalRx: Rectify and eXamine Semantic Failures in Autoformalization

FormalRx is a diagnostic evaluation framework for autoformalization built on the SCI Error Taxonomy (28 error categories). It supports alignment verdicts, error categorization, localization, and correction, and introduces FormalRx-8B and the FormalRx-Test benchmark.

Yingjia Wan, Haochen Tan, Xiao Zhu, Xinyu Zhou, Zhiwei Li, Qingsong Lv, Changxuan Sun, Jiaqi Zeng, Yi Xu, Jianqiao Lu, Yinhong Liu, Zhijiang Guo

November, 2025 EMNLP 2025 (Findings)

FaStfact: Faster, Stronger Long-Form Factuality Evaluations in LLMs

FaStfact is a multi-agent pipeline for evaluating long-form generation factuality that achieves the highest alignment with human evaluation and time/token efficiency among existing baselines. An annotated FaStfact-bench is also open-sourced.

Jianqiao Lu, Yingjia Wan, Yinya Huang, Zhengying Liu, Zhijiang Guo

June, 2024 ICLR 2025

FormalAlign: Automated Alignment Evaluation for Autoformalization

FormalAlign aims to address a long-lasting challenge in autoformalization - the absence of an effective alignment evaluation metric. It accurately assesses the semantic and logical alignment between the natural language and formal theorems.

Zhongshen Zeng, Yinhong Liu, Yingjia Wan, Jingyao Li, Pengguang Chen, Jianbo Dai, Yuxuan Yao, Rongwu Xu, Zehan Qi, Wanru Zhao, Linling Shen, Jianqiao Lu, Haochen Tan, Yukang Chen, Hao Zhang, Zhan Shi, Bailin Wang, Zhijiang Guo, Jiaya Jia

June, 2024 NeurIPS 2024

MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs

MR-BEN is a comprehensive process-based benchmark to evaluate advanced `meta-reasoning’ skills, where models are asked to locate and analyse errors in the provided CoT solutions. It comprises 5,975 multi-domain samples with annotated groundtruths.

See all publications