# An Empirical Analysis of Deep Learning for Cardinality Estimation

arXiv: https://arxiv.org/abs/1905.06425 · HTML: https://mlsanity.com/paper/1905.06425/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 15 May 2019.

## Built on by (most cited first, 8 of 8)

- [A Unified Deep Model of Learning from both Data and Queries for Cardinality Estimation](https://mlsanity.com/paper/2107.12295/) (2021)
- [A Survey on Advancing the DBMS Query Optimizer: Cardinality Estimation, Cost Model, and Plan Enumeration](https://mlsanity.com/paper/2101.01507/) (2021)
- [Efficient Deep Learning Pipelines for Accurate Cost Estimations Over Large Scale Query Workload](https://mlsanity.com/paper/2103.12465/) (2021)
- [Duet: efficient and scalable hybriD neUral rElation undersTanding](https://mlsanity.com/paper/2307.13494/) (2023)
- [HERO: Hint-Based Efficient and Reliable Query Optimizer](https://mlsanity.com/paper/2412.02372/) (2024)
- [GEqO: ML-Accelerated Semantic Equivalence Detection](https://mlsanity.com/paper/2401.01280/) (2024)
- [DistJoin: A Decoupled Join Cardinality Estimator based on Adaptive Neural Predicate Modulation](https://mlsanity.com/paper/2503.08994/) (2025)
- [Selective Use of Yannakakis' Algorithm to Improve Query Performance: Machine Learning to the Rescue](https://mlsanity.com/paper/2502.20233/) (2025)

## Abstract (arXiv)

We implement and evaluate deep learning for cardinality estimation by studying the accuracy, space and time trade-offs across several architectures. We find that simple deep learning models can learn cardinality estimations across a variety of datasets (reducing the error by 72% - 98% on average compared to PostgreSQL). In addition, we empirically evaluate the impact of injecting cardinality estimates produced by deep learning models into the PostgreSQL optimizer. In many cases, the estimates from these models lead to better query plans across all datasets, reducing the runtimes by up to 49% on select-project-join workloads. As promising as these models are, we also discuss and address some of the challenges of using them in practice.