Haifeng Wang is Baidu’s chief technology officer. A researcher in language technology, he has worked on machine translation, search and language models, and coauthored research in the ERNIE series.[1][2][4]
Start with ERNIE 2.0 →
Career
Wang earned his three computer-science degrees at Harbin Institute of Technology. He worked at Microsoft Research Asia, iSilk.com and Toshiba’s China R&D Center before joining Baidu in 2010. His own chronology dates his CTO role to 2019 and describes earlier vice-president and senior-vice-president roles.[3][2]
Contributions
ERNIE 2.0, coauthored by Wang and colleagues, treats pretraining as an expanding set of tasks. Tasks address lexical, syntactic and semantic information, and continual multi-task learning retains earlier objectives while adding new ones. The paper evaluates English and Chinese language-understanding benchmarks. The contribution is a training framework and experimental study, not proof of general reasoning across arbitrary tasks.[4]
Ideas
Wang’s research interests connect language understanding with the infrastructure needed to apply it at scale: search, translation, knowledge graphs and deep learning. His current research page also links subsequent ERNIE reports, providing a path from an early pretraining method to later model development.[2]
Tools, writing & talks
Writing, research & talks
Start here
ERNIE 2.0arxiv.orgContinual multi-task pretraining adds lexical, syntactic and semantic learning objectives.
Read the task-construction and continual-learning diagram before the benchmark tables. Familiarity with BERT-style pretraining helps; distinguish this 2019 framework from later products sharing the ERNIE name.
Co-author[4]