Cardinality estimation is a core component of query optimization, helping databases choose efficient execution plans. In recent years, learned estimators have shown promise, but most are built for a single predicate type—for example, only string LIKE clauses or only numerical range filters. This narrow focus limits their usefulness in real analytical workloads, which typically mix different kinds of filters.
To address this gap, a new arXiv paper introduces MultiCard, a deep learning approach designed for multiple attribute cardinality estimation. The authors note that analytical workloads often require estimating the number of rows that satisfy conditions on several attributes simultaneously, yet existing learned estimators rarely support this directly. MultiCard aims to fill that void by learning across different predicate types.
The abstract does not provide details on the model architecture, training data, or experimental results, so the specific advantages and trade-offs remain unclear. Still, the work points toward a more practical direction for learned cardinality estimation—one that matches the complexity of real queries rather than isolated filter types.