Computing Local Sensitivities of Counting Queries with Joins
This addresses efficiency challenges in database query analysis and differential privacy for large datasets, though it is incremental as it focuses on specific query sub-classes.
The paper tackles the NP-hard problem of computing local sensitivity for counting queries with joins, presenting algorithms that run in polynomial time for specific query classes like doubly acyclic queries, and demonstrates applications in differential privacy with orders-of-magnitude improvements.
Local sensitivity of a query Q given a database instance D, i.e. how much the output Q(D) changes when a tuple is added to D or deleted from D, has many applications including query analysis, outlier detection, and in differential privacy. However, it is NP-hard to find local sensitivity of a conjunctive query in terms of the size of the query, even for the class of acyclic queries. Although the complexity is polynomial when the query size is fixed, the naive algorithms are not efficient for large databases and queries involving multiple joins. In this paper, we present a novel approach to compute local sensitivity of counting queries involving join operations by tracking and summarizing tuple sensitivities -- the maximum change a tuple can cause in the query result when it is added or removed. We give algorithms for the sensitivity problem for full acyclic join queries using join trees, that run in polynomial time in both the size of the database and query for an interesting sub-class of queries, which we call 'doubly acyclic queries' that include path queries, and in polynomial time in combined complexity when the maximum degree in the join tree is bounded. Our algorithms can be extended to certain non-acyclic queries using generalized hypertree decompositions. We evaluate our approach experimentally, and show applications of our algorithms to obtain better results for differential privacy by orders of magnitude.