avatar

Yuxi Sun

Ph.D. Candidate in Computer Science
Hong Kong Baptist University

Biography

Hi, I am Yuxi Sun 👋, a Ph.D. candidate in the Department of Computer Science at Hong Kong Baptist University, advised by Prof. Jing Ma. I received my M.Sc. in Information Technology from City University of Hong Kong, advised by Prof. Rosa Chan, and my B.Eng. in Automation from Central South University.

My research focuses on trustworthy LLMs, spanning diagnostic evaluation, causality and LLMs, and fact-checking. I am currently extending this line of work to LLM agents.

Research Interests

  • Diagnostic Evaluation: evaluating LLMs beyond final scores, such as knowledge conflict under internal epistemic states and step-level reasoning faithfulness
  • Causality and LLMs: causal methods for assessing and improving reasoning and abstention, and foundation models built on causal priors
  • Fact-Checking and Hallucination: verifying claims against retrieved evidence and mitigating hallucination

News

2026
Sep
The Xiaomi-TabLDM technical report is now available on arXiv and GitHub 🚀. Xiaomi-TabLDM extends the foundation-model paradigm to tabular data through large-scale synthetic pretraining and in-context learning.
Aug
PAVE is accepted to EMNLP 2026 🎉. PAVE studies how LLM verifiers arbitrate between internal priors and retrieved evidence under conflict in RAG-based fact-checking.
Jul
Joined Xiaomi through the Top Talent Program as an algorithm research intern 🌟. I am working on tabular foundation models, with a focus on synthetic pretraining and scalable data priors.
Jul
FACT-E is accepted to ACL 2026 . FACT-E evaluates chain-of-thought faithfulness through causality-inspired perturbations and dependency estimation.
Feb
HOMER is accepted to ICLR 2026 . HOMER introduces a multi-role LLM framework for humor generation with conflicting-script reasoning and retrieval-augmented imagination.
2025
Dec
Our work on explainable ethical assessment appears at IJCNLP-AACL 2025 . The work improves explainable ethical judgment by reasoning over conflicting social norms behind human behaviors.
Jul
CausalAbstain is accepted to ACL 2025 . CausalAbstain uses causal reasoning to help multilingual LLMs recognize knowledge boundaries and make more reliable abstention decisions.

Publications

  1. FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning

    Yuxi Sun, Aoqi Zuo, Haotian Xie, Wei Gao, Mingming Gong, Jing Ma · ACL 2026 [Paper] [arXiv] [Code]

  2. Diagnosing LLM Arbitration Behavior over Pre-evidence Epistemic States in RAG-based Fact-Checking

    Yuxi Sun, Wenbo Shang, Wei Gao, Xin Huang, Jing Ma · EMNLP 2026 [arXiv] [Code]

  3. REFLEX: Self-Refining Explainable Fact-Checking via Verdict-Anchored Style Control

    Chuyi Kong, Wei Gao, Jing Ma, Hongzhan Lin, Yuxi Sun · ACL 2026 [Paper] [Scholar]

  4. On the Wings of Imagination: Conflicting Script-based Multi-role Framework for Humor Caption Generation

    Wenbo Shang, Yuxi Sun, Jing Ma, Xin Huang · ICLR 2026 [Paper] [arXiv] [Code]

  5. CausalAbstain: Enhancing Multilingual LLMs with Causal Reasoning for Trustworthy Abstention

    Yuxi Sun, Aoqi Zuo, Wei Gao, Jing Ma · ACL 2025 [Paper] [arXiv] [Code]

  6. SHARP: Unlocking Interactive Hallucination via Stance Transfer in Role-Playing LLMs

    Chuyi Kong, Ziyang Luo, Hongzhan Lin, Zhiyuan Fan, Yaxin Fan, Yuxi Sun, Jing Ma · ACL 2025 [Paper] [Scholar]

  7. Explainable Ethical Assessment on Human Behaviors by Generating Conflicting Social Norms

    Yuxi Sun, Wei Gao, Hongzhan Lin, Jing Ma, Wenxuan Zhang · IJCNLP-AACL 2025 [Paper] [arXiv] [Code]

Technical Report

Contact