• I am WU JUNCHAO (吴俊潮). Currently, I am a second-year Ph.D. student in Computer Science at NLP2CT Lab, University of Macau, fortunately advised by Prof. Derek F. Wong. Previously, I completed my M.S. in Data Science (Computational Lingustics) at the same lab, co-advised by Prof. Derek F. Wong and Prof. Yuan Yulin. I achineved my bachelor’s degree at Beijing Normal University, Zhuhai, supervised by Prof. JiangYing and Prof. Yangjing.
• My current research interests are Natural Language Processing, Machine Translation and Trustworthy AI. Please feel free to contact me via email!

Research Overview

My core research centers on explainable, trustworthy, and secure large language models (LLMs), with a primary focus on LLM-generated text detection and effecient model post-training. ​

• LLM-Generated Text Detection & Benchmarking: To address the opacity of AI-generated content, I released a systematic survey [CL’25] that highlights the field’s key challenges and future directions. I then developed detection frameworks leveraging grammatical [COLING’25] and representation pattern [TACL’25] differences, constructed leading benchmarks tailored to multilingual and real-world scenarios [NeurIPS’24; ACL’26] and specialized domains (e.g., modern Chinese poetry [EMNLP’25]), and organized the shared task on this topic [NLPCC’25; NLPCC’26]. ​

• Effecient and Controllable LLM Tuning & Reasoning: Focusing on efficient, controllable, and explainable LLM post-training and reasoning, I proposed a neuron-aware instruction tuning framework [ICLR’26] for efficient instruction tuning. Collaboratively, I investigated internal reasoning mechanisms of LLMs, including “aha moments” in complex problem-solving [TACL’26] and COT monitorability in LRMs [arXiv’25], to enhance model trustworthiness. ​
• Collaborative Research on Multilingual & Safety: I also engage in collaborative research on related directions, including: 1) domain-adaptive machine translation [TALLIP’26] and the exploration of LLMs as machine translation evaluators [ICML2025@AIW 2025]; 2) effecient multimodal long-document understanding [CVPR’26]; and 3) LLM safety & ethics [ACL’25; EMNLP’25; ACL’26], covering debiasing optimization, resistance to fraud/phishing inducements, implicit risk identification, and political stance mitigation.

News

• [2026-04-07] Two paper accepted by ACL 2026 Main:
  • DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection (Firsr-author): Proposes a novel multilingual LLM-generated text detection benchmark DetectRL-X, covering 8 languages. We innovatively introduces a LLM-refined writing detection task and systematically discusses the generalization ability and core challenges of detectors across languages of varying complexity and adversarial scenarios.
  • Understanding and Mitigating Political Stance Cross-topic Generalization in Large Language Models (Co-author).
• [2026-03-14] One paper accepted by CVPR 2026 Findings as Co-author: LongDocSpan: Extending LVLMs for Long Document Understanding.
• [2026-01-27] One paper accepted by ICLR 2026 as Co-first author: Neuron-Aware Data Selection in Instruction Tuning for Large Language Models. We propose the NAIT framework, which selects optimal instruction tuning data samples by evaluating the impact of IT data on LLMs performance by analyzing the similarity of neuron activation patterns between the IT dataset and the target domain capability.
• [2026-01-05] One paper accepted by ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP) as Co-author. Congratulations to Xinyi and all Co-authors!
• [2025-08-21] Two paper accepted by EMNLP 2025 Findings as Co-author. And our CL paper A Survey on LLM-generated Text Detection: Necessity, Methods, and Future Directions and TACL paper RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns will also oral present in EMNLP 2025. See you in Suzhou!!!
  • Can Large Language Models Identify Implicit Suicidal Ideation? An Empirical Evaluation.
  • Benchmarking the Detection of LLMs-Generated Modern Chinese Poetry.
• [2025-08-01] One Paper accepted by Transactions of the Association for Computational Linguistics (TACL) as Co-first author: RepreGuard: Detecting LLM-Generated Text by Revealing Hidden Representation Patterns. We propose RepreGuard, a low-overhead, highly generalizable, and interpretable detector, based on the observation that there are significant differences in neural activation patterns when LLMs process LLM-generated text versus human-written text.

Publications

2021

“The Canton Canon” Digital Library Based on Knowledge Graph - Taking the Revolutionary Archives of Canton in the Republic of China as an Example

Junchao Wu , Ying Jiang†, Xin Chen, Lingyu Guo, Xiaotong Wei and Xiaoyan Yang

Junchao Wu, Ying Jiang†, Xin Chen, Lingyu Guo, Xiaotong Wei and Xiaoyan Yang

ICEIT（2021 IEEE 10th International Conference on Educational and Information Technology）2021 2021

[PAPER] [CODE]

Best Oral Presentation Award (5/57)

Services

• Conference Reviewer: ACL ARR, ICML, ICLR, NeurIPS, AAAI, CVPR, COLM, CCL, NLPCC
• Journal Reviewer: Proc. IEEE, ACM Computing Surveys, IEEE TIFS, ACM TALLIP, ACM TIST
• Student Volunteer: MT Summit 2023
• Teaching Assistant
  • Computational Linguistics (MSc) programme (2023 Fall)
  • AHGC7315 Language and Linguistics (2023 Spring)

Experience

• Alibaba Cloud, Alibaba Group - LLM Research Intern (Feb. 2026 - Present), Supervised by Yichao Du and Longyue Wang.
• Alibaba International, Alibaba Group - LLM Research Intern (Jul. 2025 - Jan. 2026), Supervised by Yichao Du, Yefeng Liu and Longyue Wang.
• PRADA Lab, King Abdullah University of Science and Technology (KAUST) - Visiting Research Intern (Mar. 2025 - Jun. 2025), Supervised by Prof. Di Wang.

Professional skills

• English Level: IELTS(6.5), CET-6(507)
• Programming Languages: C, Python, Java, JavaScript, SQL, Bash
• Development Frameworks: SpringBoot, React.js, Flask
• Deep Learning Tools: Scikit-learn, PyTorch, Fairseq, Transformers
• DB Tools: Neo4j, MySQL, Oracle DB
• Other Tools: WebLogic, Apache Ant, Jena, Git

Others

• 🎓 If you are also interested in natural language processing and machine translation, you can follow my lab and lab members: NLP2CT LAB, I love them.
• ✨Zhan Runzhe is my current advisor, who focus on machine translation and LLMs. He is a brilliant scientist and one of the nicest people I have ever met.
• 🌈Chen Xin is one of my best friends, who is presently a PhD student at Nanjing University and is also committed to NLP research., he has many interesting dreams.
• ❤️ He Qiufeng, the love of my life and the best gift life has given me. A brilliant PhD candidate in civil engineering at Shenzhen University, and we dream to write the future of AI + Civil Engineering together. With her,everything matters, and everything awaits.