Notice

CSE will be holding a special seminar with outstanding researchers.
Even if you are busy, we hope you will attend and have a special time with outstanding researchers.

1. Date & Time : December 23rd(Mon), 14:00
2. Venue : 104동 E204호
3. Speaker: Ph.D Kyoungmin Kim (DIAS lab @ EPFL)
4. Research areas: DB + AI
5. Talk title: Cardinality Estimation using Sampling, Synopsis, and Deep Learning
6. Abstract: Cardinality estimation lies at the core of query optimization in databases management systems (DBMSes). Given an SQL query, where an optimizer seeks for the best plan to execute, it estimates the costs of candidate plans based on the estimated cardinalities (the number of output tuples) of subplans. Therefore, a cardinality estimator should be accurate enough to find the best plan and efficient enough to be called for thousands of subplans without incurring a large overhead. For decades, sampling and synopsis have been two major approaches for cardinality estimation, relying on ad-hoc heuristics and assumptions, e.g., independence between columns and uniform distribution of values. While they are extremely fast in general, improving their accuracy, without severely degrading efficiency, has been remained as a non-trivial, challenging task. This presentation introduces new techniques based on sampling, synopsis, and deep learning for boosting the performance of cardinality estimation for graph and relational databases, especially for select-join-project (SPJ) queries.