Wang Xiaozhi

Xiaozhi WANG is an Assistant Professor at Shenzhen International Graduate School, Tsinghua University. He obtained his Bachelor's and Ph.D. degrees from Department of Computer Science and Technology at Tsinghua University. His research focuses on large language models, natural language processing, and knowledge engineering, with a particular emphasis on mechanistic interpretability of large models and mechanism-guided improvements in model architecture, training, and evaluation.

Xiaozhi WANG has published multiple papers in top artificial intelligence conferences and journals such as ACL, EMNLP, and ICLR, which have garnered over 4,000 citations. He has also served as an area chair for several important conferences like ACL and EMNLP. His work received EMNLP Outstanding Paper Award and was listed as ESI Highly Cited Papers.

The research group continuously seeks highly motivated Ph.D., Master, postdoc, and research intern. Welcome to reach out.

Academic homepage: https://bakser.github.io

2020.8-2025.6 Department of Computer Science and Technology, Tsinghua University, Ph.D.

2016.8-2020.6 Department of Computer Science and Technology, Tsinghua University, Bachelor

2025.8-Present Institute of Data and Information, Shenzhen International Graduate School, Tsinghua University, Assistant Professor

2024.2-2024.8 University of Illinois Urbana-Champaign, Visiting Scholar

2019.7-2019.9 Montreal Institute for Learning Algorithm, Visiting Scholar

Area Chair: ACL Rolling Review, ACL/EMNLP/NAACL 2024, ACL/EMNLP 2025

Workshop Co-organizer: AAAI/ACL 2025 Workshop Towards Knowledgeable Foundation Models

Reviewer/Program Committee Member: AAAI, IJCAI, COLING, SIGIR, ACL, EMNLP, NeurIPS, IEEE/ACM TASLP, IEEE TKDE

Xiaozhi WANG's research areas include large language models, natural language processing, and knowledge engineering. His primary research interests encompass:

• Mechanistic Interpretability of Large Models: Understanding the inner workings of large models to explain their behaviors while exploring the scientific principles behind the emergence of artificial intelligence.

• Scientific Improvements in Large Models: Enhancing the architecture, training, and evaluation of large models based on the mechanistic understandings, such as exploring sparsely activated architectures, guiding data selection and model evaluation with internal states, thereby fundamentally improving the knowledge, efficiency, safety, and autonomy of large models.

He has also conducted research in knowledge acquisition and knowledge-enhanced large models. In the above areas, his representative achievements include techniques for locating skill neurons in large models, comprehensive evaluation systems for large models' knowledge understanding, and knowledge-enhanced technologies. He has published over ten papers as (co-)first author in top-tier international conferences and journals such as ACL, EMNLP, and ICLR, and has been granted three national invention patents. His received over 4,500 citations and honors such as ESI Highly Cited Paper and EMNLP Outstanding Paper Award.

Please refer to https://bakser.github.io/publications for the complete list.

[1] Wang X, Gao T, Zhu Z, et al. KEPLER: A unified model for knowledge embedding and pre-trained language representation[J]. Transactions of the Association for Computational Linguistics (TACL), 2021, 9: 176-194.

[2] Wang X, Wang Z, Han X, et al. MAVEN: A Massive General Domain Event Detection Dataset[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020: 1652-1671.

[3] Wang X, Peng H, Guan Y, et al. MAVEN-ARG: Completing the Puzzle of All-in-One Event Understanding Dataset with Event Argument Annotation[C]//Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL). 2024: 4072-4091.

[4] Wang X*, Chen Y*, Ding N, et al. MAVEN-ERE: A Unified Large-scale Dataset for Event Coreference, Temporal, Causal, and Subevent Relation Extraction[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). 2022: 926-941.

[5] Wang X*, Wen K*, Zhang Z, et al. Finding Skill Neurons in Pre-trained Transformer-based Language Models[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). 2022: 11132-11152.

[6] Wang Z*, Wang X*, Han X, et al. CLEVE: Contrastive Pre-training for Event Extraction[C]//Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL). 2021: 6283-6297.

[7] Su Y*, Wang X*, Qin Y, et al. On Transferability of Prompt Tuning for Natural Language Processing[C]//Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics (NAACL). 2022: 3949-3969.

[8] Peng H*, Wang X*, Hu S, et al. COPEN: Probing Conceptual Knowledge in Pretrained Language Models[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). 2022: 5015-5035.

[9] Peng H*, Wang X*, Yao F, et al. OmniEvent: A Comprehensive, Fair, and Easy-toUse Toolkit for Event Understanding[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP): System Demonstrations. 2023: 508-517.

[10] Peng H*, Wang X*, Yao F*, et al. The Devil is in the Details: On the Pitfalls of Event Extraction Evaluation[C]//Findings of the Association for Computational Linguistics: ACL 2023. 2023: 9206-9227.

[11] Yu J*, Wang X*, Tu S*, et al. KoLA: Carefully Benchmarking World Knowledge of Large Language Models[C]//Proceedings of the International Conference on Learning Representations (ICLR). 2024.

[12] Qin Y*, Wang X*, Su Y, et al. Exploring Universal Intrinsic Task Subspace for Few-Shot Learning via Prompt Tuning[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024, 32: 3631-3643.

Excellent PhD Graduate of DCST, Tsinghua University, 2025

China National Scholarship, 2024

EMNLP Outstanding Paper Award, 2023