Abducing Relations in Continuous Spaces

Authors: Taisuke Sato, Katsumi Inoue, Chiaki Sakama

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We next test our linear algebraic approach using artificial data and real data. ... We conduct an experiment for non-recursive abduction with artificial data. ... We use FB15k, a standard knowledge graph comprised of triples ... We conduct an experiment with artificial data to test the above approach. ... We then conduct a similar experiment with real data. We choose five network graphs from the Koblenz Network Collection
Researcher Affiliation Academia Taisuke Sato1, Katsumi Inoue2, Chiaki Sakama3 1 AI research center AIST, Japan 2 National Institute of Informatics, Japan 3 Wakayama University, Japan
Pseudocode No The paper describes its methods using mathematical equations and text but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating the availability of open-source code for the methodology described.
Open Datasets Yes We use FB15k, a standard knowledge graph comprised of triples (subject, relation, object) in RDF format for 1,345 binary relations and 14,951 entities [Bordes et al., 2013]. ... We choose five network graphs from the Koblenz Network Collection [Kunegis, 2013]
Dataset Splits Yes The triples comprising a relation in FB15k are divided into training, validation and test sets [Bordes et al., 2013]. We use training sets for rule and relation discovery.
Hardware Specification Yes All experiments in this paper are carried out by GNU Octave 4.0.0 on a PC with Intel(R) Core(TM) i7-3770@3.40GHz CPU, 28GB memory.
Software Dependencies Yes All experiments in this paper are carried out by GNU Octave 4.0.0 on a PC with Intel(R) Core(TM) i7-3770@3.40GHz CPU, 28GB memory.
Experiment Setup Yes In the experiment, we set n = 10^4, create two n n random adjacency matrices R1 and R2 ... We repeat this process for λ = 1 and pe {10^-2,10^-3,10^-4,10^-5}. ... We set n = 10^4 and θ = 10^-4. Given pe, we generate an n n random matrix R1 encoding D(n,pe) and compute its transitive closure R2.