A Unified Approach to Incremental View Maintenance of In-Database Analytics

In this talk, we study the problem of incremental computation of analytical tasks over joins. We introduce a principled incremental view maintenance (IVM) mechanism that reduces the maintenance of the given task to the maintenance of a hierarchy of increasingly simpler views. The views are functions mapping keys, which are tuples of input data values, to payloads, which are elements from a task-specific ring. The computation over the keys is the same for all tasks, while the computation over the payloads depends on the task. Our IVM approach achieves efficiency by factorizing the computation of the keys, payloads, and updates. We demonstrate our technique on a variety of analytical tasks, such as gradient computation for learning regression models, linear algebra operations, and factorized evaluation of conjunctive queries, and experimentally show that it can achieve orders of magnitude better performance and lower memory consumption than existing state-of-the-art IVM techniques.