Xin Liu (刘鑫)

• I am a Ph.D. candidate in Computer Science & Engineering at the University of Michigan and a member of the LAUNCH lab, advised by Prof. Lu Wang. My research builds trustworthy and efficient large language models — spanning factuality, confidence calibration, reasoning and efficient inference, long-form evaluation, and knowledge distillation / post-training, and, most recently, interpretability and safety. I enjoy building large-scale training and evaluation pipelines (for 1B–70B models) and turning research into reliable, deployable model behavior. Previously, I received my master’s and bachelor’s degrees at Xiamen University, where I was advised by Prof. Jinsong Su.

• I am on the job market for research scientist positions, graduating in December 2026 — please feel free to reach out!

• Email: Please arrange edu @ . liuxincs umich to get my email address.

• Google Scholar, DBLP, LinkedIn

News

• [May 2026] Joined ByteDance Seed as a Research Intern, working on LLM safety and interpretability.
• [Apr 2026] Paper Logit Arithmetic Elicits Long Reasoning Capabilities Without Training is accepted to Findings of ACL 2026.
• [Aug 2025] Two papers — VeriFact: Enhancing Long-Form Factuality Evaluation with Refined Fact Extraction and Reference Facts and Answer Convergence as a Signal for Early Stopping in Reasoning — are accepted to EMNLP 2025.
• [May 2025] Started a GenAI Research internship at Zoom.
• [Sep 2024] Paper Enhancing Language Model Factuality via Activation-Based Confidence Calibration and Guided Decoding is accepted to EMNLP 2024.
• [May 2024] Started a Research internship at Tencent America.
• [May 2024] Paper Enhanced Language Model Truthfulness with Learnable Intervention and Uncertainty Expression is accepted as a Findings paper of ACL 2024.
• [Jan 2024] Paper LitCab: Lightweight Language Model Calibration over Short- and Long-form Responses is accepted to ICLR 2024.
• [May 2023] Paper BOLT: Fast Energy-based Controlled Text Generation with Tunable Biases is accepted to ACL 2023.
• [Aug 2022] Joined the LAUNCH lab at the University of Michigan as a Ph.D. student, advised by Prof. Lu Wang.
• [Mar 2022] Nominated as the Alibaba Group Outstanding Science Research Intern (13 nominations out of 1,000+ candidates).
• [Dec 2021] Paper KGR^4: Retrieval, Retrospect, Refine and Rethink for Commonsense Generation is accepted by AAAI 2022.
• [May 2021] Our model KGR^4 achieves 1st place in CommonGen leaderboard.

Education

• [Aug 2022-] Ph.D. in Computer Science, University of Michigan. Advisor: Prof. Lu Wang.
• [Aug 2019-Jun 2022] M.S. in Software Engineering, Xiamen University. Advisor: Prof. Jinsong Su.
• [Aug 2015-Jun 2019] B.S. in Software Engineering, Xiamen University.

Internship Experience

• [May 2026-present] Research Intern at ByteDance Seed (LLM Safety & Interpretability).
• [May 2025-Aug 2025] GenAI Research Intern at Zoom Communications.
• [May 2024-Aug 2024] Research Intern at Tencent America.
• [Oct 2020-Jun 2022] Research Intern at Alibaba DAMO Academy. Advisors: Dr. Baosong Yang and Dr. Dayiheng Liu.
• [Mar 2019-Feb 2020] Research Intern at Baidu NLP. Advisor: Dr. Kai Liu.

Publications

(* denotes equal contribution; bold denotes me)

• Xin Liu, Simin Ma, Shujian Liu, Song Wang, Sathish Reddy Indurthi, Haoyun Deng, Lu Wang, Kaiqiang Song. Understanding Knowledge Distillation in Post-Training: When It Helps and When It Fails. Preprint, 2026 (under review)
• Xin Liu, Lu Wang. Think Through Uncertainty: Improving Long-Form Generation Factuality via Reasoning Calibration. Preprint, 2026
• Yunxiang Zhang, Muhammad Khalifa, Lechen Zhang, Xin Liu, Ayoung Lee, Xinliang Frederick Zhang, Farima Fatahi Bayat, Lu Wang. Logit Arithmetic Elicits Long Reasoning Capabilities Without Training. In Findings of ACL 2026
• Xin Liu, Lu Wang. Answer Convergence as a Signal for Early Stopping in Reasoning. In EMNLP 2025 (short)
• Xin Liu, Lechen Zhang, Sheza Munir, Yiyang Gu, Lu Wang. VeriFact: Enhancing Long-Form Factuality Evaluation with Refined Fact Extraction and Reference Facts. In EMNLP 2025
• Xiaoyue Wang, Lijie Wang, Xin Liu, Suhang Wu, Jinsong Su, Hua Wu. A Simple yet Effective Self-Debiasing Framework for Transformer Models. Artificial Intelligence (Journal), 2025
• Xin Liu, Farima Fatahi Bayat, Lu Wang. Enhancing Language Model Factuality via Activation-Based Confidence Calibration and Guided Decoding. In EMNLP 2024
• Farima Fatahi Bayat, Xin Liu, H. V. Jagadish, Lu Wang. Enhanced Language Model Truthfulness with Learnable Intervention and Uncertainty Expression. In Findings of ACL 2024
• Xin Liu, Muhammad Khalifa, Lu Wang. LitCab: Lightweight Language Model Calibration over Short- and Long-form Responses. In ICLR 2024
• Xiaoyue Wang, Xin Liu, Lijie Wang, Yaoxiang Wang, Jinsong Su, Hua Wu. IBADR: An Iterative Bias-Aware Dataset Refinement Framework for Debiasing NLU Models. In EMNLP 2023
• Xin Liu, Muhammad Khalifa, Lu Wang. BOLT: Fast Energy-based Controlled Text Generation with Tunable Biases. In ACL 2023
• Xiaoyue Wang, Linfeng Song, Xin Liu, Chulun Zhou, Hualin Zeng, Jinsong Su. Getting the Most out of Simile Recognition. In Findings of EMNLP 2022
• Xin Liu, Dayiheng Liu, Baosong Yang, Haibo Zhang, Junwei Ding, Wenqing Yao, Weihua Luo, Haiying Zhang, Jinsong Su. KGR^4: Retrieval, Retrospect, Refine and Rethink for Commonsense Generation. In AAAI 2022
• Xin Liu, Baosong Yang, Dayiheng Liu, Haibo Zhang, Weihua Luo, Min Zhang, Haiying Zhang, Jinsong Su. Bridging Subword Gaps in Pretrain-Finetune Paradigm for Natural Language Generation. In ACL 2021
• Xin Liu*, Kai Liu*, Xiang Li*, Jinsong Su, Yubin Ge, Bin Wang, Jiebo Luo. An Iterative Multi-Source Mutual Knowledge Transfer Framework for Machine Reading Comprehension. In IJCAI 2020
• Kai Liu*, Xin Liu*, An Yang, Jing Liu, Jinsong Su, Sujian Li, Qiaoqiao She. A Robust Adversarial Training Approach to Machine Reading Comprehension. In AAAI 2020
• Qian Lin, Huating Wen, Jing Yang, Xin Liu, Huan Lin, Hongji Wang, Jinsong Su. The construction of internal hierarchical structure corpus for Chinese words. Journal of Xiamen University (Natural Science Edition)