Learned / ML-based Query Optimization
Parent: Machine Learning · Topic entry · 15 branches
A published reference is not available for this topic yet.
Children
- Why cardinality estimation is the optimizer's hardest sub-problem (frontier)
- Query-driven learned cardinality estimation (MSCN) (frontier)
- Data-driven learned cardinality estimation (DeepDB, Naru, NeuroCard) (frontier)
- Pessimistic/bound-based estimation (AGM/PANDA/SafeBound/LpBound) (frontier)
- Learned cost models (QPPNet, tree-LSTM, tree-CNN, zero-shot) (frontier)
- Neo end-to-end learned optimizer (frontier)
- RL join ordering (ReJOIN, DQ, Balsa) (frontier)
- Replacement-vs-steering spectrum (Bao, Lero, AutoSteer) (frontier)
- Instance-optimized / self-driving databases (Peloton/NoisePage) (frontier)
- OtterTune knob tuning and shutdown (frontier)
- Learned indexes (RMI) adjacent (frontier)
- Benchmarks (JOB/IMDB, STATS-CEB, JOB-Complex) (frontier)
- q-error vs P-Error / plan regret (frontier)
- When learned optimization fails (frontier)
- Production status as of 2026 (QO-Advisor shutdown, ByteCard) (frontier)
Frontier under this node: Benchmarks (JOB/IMDB, STATS-CEB, JOB-Complex), Data-driven learned cardinality estimation (DeepDB, Naru, NeuroCard), Instance-optimized / self-driving databases (Peloton/NoisePage), Learned cost models (QPPNet, tree-LSTM, tree-CNN, zero-shot), Learned indexes (RMI) adjacent, Neo end-to-end learned optimizer, OtterTune knob tuning and shutdown, Pessimistic/bound-based estimation (AGM/PANDA/SafeBound/LpBound), Production status as of 2026 (QO-Advisor shutdown, ByteCard), Query-driven learned cardinality estimation (MSCN), RL join ordering (ReJOIN, DQ, Balsa), Replacement-vs-steering spectrum (Bao, Lero, AutoSteer), When learned optimization fails, Why cardinality estimation is the optimizer's hardest sub-problem, q-error vs P-Error / plan regret