BIFOLD Talk: Attention-based Learned Cardinality Estimator for SPJ queries.

Abstract:

Pengfei Li will introduce ALECE, an Attention-based Learned Cardinality Estimator for SPJ queries. The key idea is to capture the implicit relationships between queries and dynamic underlying data through two attention-based modules built on carefully designed data and query featurizations. Specifically, the data-encoder module learns organic aggregations over database attributes to implicitly model correlations among them, while the query-analyzer module connects query features with these data aggregations to estimate query cardinalities. I will then briefly discuss how ALECE can be extended into an optimizer-facing interface that models the uncertainty of cardinality estimates, thereby improving the robustness of query optimizers.

Bio:

Pengfei Li is currently a postdoctoral researcher at Roskilde University, Denmark. He received his Ph.D. in Computer Science and Technology from Zhejiang University, China. Before joining Roskilde University, he worked as a Senior Algorithm Engineer at Alibaba Cloud. His research interests include database management, query optimization, and machine learning. His recent work focuses on AI-enhanced data management and learned query optimization, with publications in leading venues such as SIGMOD, VLDB, and NeurIPS.