Filtered ANN as a Phase Transition: When Selectivity-Estimation Error Causes Plan Regret
Two new research papers explore the nuances of query optimization in large-scale data systems, focusing on how estimation errors impact performance. The first paper, "Filtered ANN as a Phase Transition," analyzes approximate nearest neighbor searches and identifies critical error regions where performance degrades significantly. The second paper, "When Does q-error Predict Plan Regret?", investigates cardinality estimation and proposes new metrics, like ACS-infinity, that better predict query plan quality than traditional q-error, especially for complex learned estimators. AI