February 16, 2023 Thursday
Recommender system (Recsys) has long been viewed as a subfield of machine learning. During my relatively short career in Recsys, this viewpoint has constantly led me to local and myopic solutions. Building a competitive Recsys not only require supervised / semi-supervise / unsupervised learning, causal inference, reinforcement learning, but also high-quality ML operations and substantial human-in-the-loop development. In this talk, I will present a more foundational view on modern Recsys and share high-level insights that challenge the conventional ML approach to Recsys.
Staff AI Engineer
LinkedIn Network AIteam
Da XU is currently a Staff AI Engineer at LinkedIn Network AI team. Prior to joining LinkedIn, he was a Manager of Machine Learning at the Search & Recommendation team in Walmart Labs. Da was with the Statistics department of UC Berkeley before joining the industry, and his primary research effort goes into establishing foundations for industrial machine learning in pattern recognition, causal inference, and decision making. Da’s research and application works have been published in such as NeurIPS, ICML, ICLR, AAAI, KDD, WWW, and WSDM. He organized the 1st and 2nd Workshop on Decision Making for IR and Recsys, KDD Tutorial on Theoretical Foundation for IR and Recsys, and serve as the co-editor of TORS Special Issue on Causal Inference in Recsys.