Lifting Relational MAP-LPs Using Cluster Signatures

Authors: Udi Apsel, Kristian Kersting, Martin Mladenov

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our new approach with several empirical results, putting special emphasis on the challenging class of transitive relational models. From the empirical results, we highlight several lifting of the SA hierarchy up to level 6, in non-trivial models.
Researcher Affiliation Academia Udi Apsel Computer Science Department Ben Gurion University of The Negev, Israel apsel@cs.bgu.ac.il Kristian Kersting and Martin Mladenov Computer Science Department TU Dortmund University, Germany {kristian.kersting, martin.mladenov}@cs.tu-dortmund.de
Pseudocode Yes Algorithm 1: GENERATECOMPACTMAPLPk
Open Source Code No The paper mentions using the 'nauty (Mc Kay and Piperno 2014) software package' but does not state that the code for their own methodology is open-source or provide a link.
Open Datasets No The paper defines a 'transitive model' using a parfactor and mentions setting 'table entries were set at random', indicating synthetic data generation rather than the use of a publicly accessible dataset with explicit access information.
Dataset Splits No The paper does not specify dataset splits (e.g., train/validation/test percentages or counts) or reference standard predefined splits.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments.
Software Dependencies No The paper mentions using the 'nauty (Mc Kay and Piperno 2014) software package' and 'the GNU Linear Programming Kit (GLPK) simplex solver,' but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We follow Algorithm 1, as follows. (1) Canonical clusters are obtained by generating all non-isomorphic instances of directed graphs with up to k edges, using the nauty (Mc Kay and Piperno 2014) software package, where all graphs assume a canonically labeled form. (2) A canonical cluster of size 3 consisting of nodes u, v, w and edges (u, v), (u, w), (v, w), is matched with the parfactor, and a linear expression involving its respective ยต variables is added to the objective. (3) For each canonical cluster of size d, we obtain subset clusters of size d 1 for which local constraints are added, by removing edges (one each time) from the d size cluster s graph and obtaining a matching canonical labeling (nauty).