* Technical Leader in the AI department Cloud BU: Investigate cloud-edge interactions for AI-related applications, including deep neural network compression, workload assignment on cloud and edge, model management as a service, etc.* Document Classification on Defects in Manufacture Systems: Design algorithms for document classification, including word segmentation, keywords extraction, ensemble classifier；Apply on a real dataset with 20 categories and more than 400 sub-categories, the classification accuracy is 99.1% on categories and 90.7% on sub-categories.* Technical Assistant of the Chief Data Scientist: Support the architecture design of AI services on Huawei public cloud; Organize collaboration projects with top univeristies on innovation topics: including one-shot learning, cloud-edge training, next-generation knowledge graph etc.

* State Error Detection on Machine Generated Data: Propose a framework of error detection applied to machine-generated data; Investigate the glitches/errors that cannot be recognized correctly by classical statistical outlier detection methods; Design a hierarchical segment approach to label correct states on data.

* Metadata Management of NoSQL Data Store: Propose a framework of metadata management on NoSQL data stores for supporting applications as data integration, data exploration, etc.; Apply a hierarchical method to group records with isomorphic schemas with a processing speed as 30,000 records/second; Propose an effective summary to represent schemas for dealing with flexible data model challenges in NoSQL data stores.* Graph Search and Analysis for Application Performance Management(APM): Propose to use graph search technology in APM; Implement recommendation for root cause analysis which changes user experiences in the bottleneck detection significantly; Design an automated method to make abnormal detection without any pre-defined threshold.

* Data Analysis Acceleration by Modern Hardware: Accelerate sub-sequence similarity search with FPGA (100x speedup of the state of art); Accelerate frequent item counting with FPGA (3x speed up of the state of art and save half of LUT); Propose a framework to accelerate time series analysis algorithms by FPGA and present a user-friendly interface for users to convert an algorithm to its FPGA design.* Spatio-Temporal Data Mining: Traffic flow detection and position users' meaningful location based on mobility data; Spatio-Temporal data mining for abnormal trajectory detection