When Plans Change Answers: Formalizing Cost-Accuracy Optimization for Semantic Queries
Semantic query plans can change both runtime and answers, so the paper defines optimization around cost and output quality together.
Kim frames semantic operators as model decisions whose calibrated confidence can be converted into expected error. The paper weights those errors by their contribution to query output, including join fan-out, to estimate plan quality without labeled data. It also shows where familiar rewrites break: selection pushdown is not quality-sound when escalation bands are calibrated on the plan’s own candidates. The work is formal and simulation-based; evaluation on real engines is left for future work. ArXiv · AI/CL/LG's note
Kim frames semantic operators as model decisions whose calibrated confidence can be converted into expected error. The paper weights those errors by their contribution to query output, including join fan-out, to estimate plan quality without labeled data. It also shows where familiar rewrites break: selection pushdown is not quality-sound when escalation bands are calibrated on the plan’s own candidates. The work is formal and simulation-based; evaluation on real engines is left for future work. ArXiv · AI/CL/LG's note
score 4