VLog: A Column-oriented Rule-Based Engine

From International Center for Computational Logic

VLog: A Column-oriented Rule-Based Engine

Talk by Jacopo Urbani
In this talk, I will first introduce our young research group in Amsterdam which is concerned with the design and implementation of efficient large-scale systems for AI. Then, I will present VLog, one of such a systems which is designed to execute Datalog reasoning and is developed in collaboration with TUD. VLog adopts the distinctive approach of computing the Datalog materialisation by storing the inference with a columnar layout. It has been shown that columnar relational technology can execute efficiently analytical SQL queries, but it deals poorly with updates. For Datalog queries, however, the problem of updates can be avoided altogether by working in append-only mode and by producing the inferences one "set-at-a-time” instead of one "fact-at-a-time”. I will first introduce the core design principles of VLog, and describe how its approach leads to good data compression and allows the implementation of techniques for avoiding duplicates. Then, I will present some experiments and give a short demo that illustrates its potential on large knowledge bases. In the last part of the talk, I will briefly report on some ongoing work on extending VLog to handle existentially quantified rules, statistical inference, and on a novel machine learning technique for improving query-driven evaluation.