Senior Expert at AI Lab, Huawei Hong Kong Research Center, Hong Kong SAR
Email: Jiang.Xin [at] huawei.com
Address: Bio-informatics Center, Hong Kong Science Park, New Territories, Hong Kong SAR
Bio
Xin Jiang is a Senior Expert at the AI Lab of Huawei Hong Kong Research Center and the manager of the large-model team in the Small and Agile Commando Team for Huawei Ascend AI Computing. He joined Huawei in 2013, worked as a researcher and project manager at Noah’s Ark Lab, and served as Director of the Speech and Semantics Lab from 2017 to 2023. Before that, he worked as a Senior Engineer at Baidu from 2011 to 2013 and at NetEase Youdao from 2009 to 2011. He received his B.S. in Mechanics and his Ph.D. in Applied Mathematics from Peking University.
His research areas include large language models, natural language processing, machine learning, and information retrieval. He was one of the technical leaders in the research and development of Huawei’s PanGu language models on Ascend NPUs from 2020 to early 2025, including PanGu-Alpha, PanGu-Sigma, and subsequent PanGu-series models. He has published papers and technical reports in leading AI, NLP, and ML conferences and journals.
Research Interests
- Large language models: architecture design, pre-training, optimization, training stability, agentic post-training
- Model compression, quantization, distillation and efficient training/inference
- Natural language understanding and generation, deep learning, machine learning, information retrieval
Work Experience
Huawei Hong Kong Research Center
Senior Expert at AI Lab, March 2025 - Present, Hong Kong
- Work on large language models and enterprise applications, with emphasis on NPU-efficient model design, pre-training, supervised fine-tuning, and reinforcement learning for reasoning and agentic tasks.
- Support industrial customers in training domain-enhanced models on Ascend NPU clusters and deploying them in business scenarios.
Huawei Noah’s Ark Lab
Principal Researcher, November 2023 - February 2025, Hong Kong
- Led and contributed to PanGu-series model research and development, mainly on large-scale MoE model pre-training on NPU clusters, with systematic studies on MoE architecture design and training strategies, including model initialization, expert granularity, load-balancing control, and learning-rate/batch-size scheduling.
- Led research projects on next-generation foundation-model architectures, including hybrid sparse/linear attention, LLM training stability, diffusion language models, optimizers, and positional encoding.
Director of Speech and Semantics Lab, July 2017 - October 2023, Hong Kong
- Directed research projects in pre-trained large models, speech and natural language dialogue systems, pretrained model compression, etc.
- Co-led (with Peng Cheng Laboratory) the development of PanGu-Alpha, a 200B-parameter autoregressive language model, and PanGu-Sigma, a trillion-parameter sparse MoE language model, both on Huawei full-stack infrastructure.
Senior Researcher, August 2013 - July 2017, Hong Kong
- Conducted research in deep learning for natural language processing and information retrieval, particularly on paraphrase generation, question answering, semantic matching, and ranking.
Baidu, Inc.
Senior Engineer, September 2011 - July 2013, Beijing
- Improved search quality of Baidu.com within a learning-to-rank framework, working on ranking algorithms, query understanding, click models, and search relevance technologies.
NetEase Youdao
Senior Engineer, July 2009 - September 2011, Beijing
- Improved search quality of Youdao.com, conducting research and development on ranking algorithms, query rewriting, PageRank optimization, and PLSI topic modeling.
Microsoft Research Asia
Intern, November 2007 - June 2008, Beijing
- Conducted research on key-phrase extraction using learning-to-rank techniques.
Education
- Peking University, Ph.D. in Applied Mathematics, supervised by Prof. Ming Jiang, Prof. Bin Yu and Dr. Hang Li, 2004 - 2009.
- Peking University, B.S. in Theoretical and Applied Mechanics, Minor in Computer Science, 2000 - 2004.
Publication List
See Google Scholar.
Awards and Honors
- ACL 2023 Outstanding Paper Award, for work on CAME: Confidence-guided Adaptive Memory Efficient Optimization.
- ACL 2022 Outstanding Paper Award, for work on compression of generative pre-trained language models via quantization.
- First Prize of the State Natural Science Award, Higher Education Outstanding Scientific Research Achievement Award (Science and Technology), Ministry of Education, China, 2022.
- ACL 2019 Best Paper Nomination, for work on Decomposable Neural Paraphrase Generation.
- Microsoft Fellowship, 2006.
Professional Service
- Program Committee member for ACL, EMNLP, NAACL, ICLR, NeurIPS, ICML, CVPR, AAAI, IJCAI, WWW, SIGIR, KDD, WSDM, and CIKM.