Lineage Trigger Graphs

Probabilistic Reasoning Under the Possible World Semantics

C++ engine to support sound and complete query answering over probabilistic knowledge graphs using Datalog rules under the distribution semantics. Lineage Trigger Graphs (LTGs) is an extension of Trigger Graphs. LTGs outperforms by several orders of magnitude engines, such as ProbLog2, vProbLog, and Scallop, both in terms of runtime and memory overhead, even without approximations. The technology behind LTGs extends the notion of provenance circuits. LTGs can be applied to visual question answering, outperforming LXMERT and RVC on the VQAR benchmark. Like GLog, LTGs can be deployed on mobile phones under the Android NDK for reasoning that runs exclusively on-device.

Instructions and scripts for reproducing the experiments in SIGMOD 2023 can be found here.

Repository

GLog

Relevant publications

2020

  1. AAAI
    Beyond the Grounding Bottleneck: Datalog Techniques for Inference in Probabilistic Logic Programs
    Efthymia Tsamoura, Vı́ctor Gutiérrez-Basulto, and Angelika Kimmig
    In The Thirty-Fourth AAAI Conference on Artificial Intelligence, New York, NY, USA, February 7-12, 2020
    DOI Code

2023

  1. PACMMOD
    Probabilistic Reasoning at Scale: Trigger Graphs to the Rescue
    Efthymia Tsamoura, Jaehun Lee, and Jacopo Urbani
    Proceedings of the ACM on Management of Data, 2023
    DOI arXiv Code