Built a team from 0 to 40 to revolutionize the music and entertainment industry using innovative technology. Documentary filmed by the Discovery Channel in China by Numbers series.We search for new business growth point in the industry where technology is the bottleneck. We solve the technology bottleneck by efficiently developing innovative techniques with innovative team management.We have developed PDM, an innovative algorithm which predicts the values of P/UGC based on multimedia contents. With PDM, contents published every single day will be analyzed and talents are discovered instantly. In 2020, PDM helped 18 songs to become top hit songs, including 《微微》《燕无歇》《踏山河》. It has been a breakthrough where we use technology to help talents to realize their value in a fair and efficient way. A documentary about PDM filmed by the Discovery Channel has been released in the China by Numbers series in 2022.In the meantime, we are developing innovative techniques to incubate multiple billion-level confidential new business.

We pursue general AI via studying the challenging Real Time Strategy game AI. Honour of Kings is currently the most popular RTS (5v5 MOBA) game in China with 80 million DAU. I have led the team develop innovative deep learning algorithms in both macro strategies and micro combat for full games. I have also led the team build Damo (达摩), one of the most difficult heroes. We made our debut in KPL final and defeated a top 1% human team which was formed with pre-pro players and casters. Our Damo is widely recognized by audience and the community as one of the best Damos among all (国服水平).I have published a paper "Hierarchical Macro Strategy Model for MOBA Game AI" as the first author in AAAI 2019 (oral presentation), introducing macro strategy level modeling. For more information, please refer to the attached press release.

Project owner of Duer drama recommender system and Duer singing voice synthesis system.

Graph algorithms have attracted high attention in Tencent Recommendation Group due to its fine generality. During this period, I independently studied Spark GraphX and developed a recommender system using YouTube’s video recommendation algorithm. The system has successfully sped up the model iteration to less than one hour.