Explanation-Based Approximate Weighted Model Counting for Probabilistic Logics

Authors: Joris Renkens, Angelika Kimmig, Guy Van den Broeck, Luc De Raedt

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

Reproducibility Variable Result LLM Response
Research Type Experimental An empirical evaluation on probabilistic logic programs shows that our approach is effective in many cases that are currently beyond the reach of exact methods. [...] Our technique is empirically evaluated on a number of benchmark problems. The experiments provide evidence that it is effective for probabilistic languages working under Sato s distribution semantics (Sato 1995)...
Researcher Affiliation Academia Joris Renkens, Angelika Kimmig, Guy Van den Broeck, Luc De Raedt KU Leuven Celestijnenlaan 200A 3001 Heverlee Belgium
Pseudocode Yes Algorithm 1: Computing a lower bound on WMC( q) given a time budget max Time. [...] Algorithm 2: Computing lower as well as upper bounds in the PLP setting given a time budget max Time.
Open Source Code No The paper mentions using the third-party WPMS solver WPM1 (Ans otegui, Bonet, and Levy 2009) but does not provide any link or statement about the availability of the authors' own implementation code.
Open Datasets Yes Our comparison uses two datasets. The first is WEBKB1, where university webpages have to be tagged with classes. [...] 1http://www.cs.cmu.edu/ webkb/ [...] The second dataset is the biological application of (Ourfali et al. 2007)... We refer to this dataset as GENES.
Dataset Splits No The paper discusses the datasets and queries used for evaluation but does not specify how they were partitioned into distinct training, validation, and test sets or refer to predefined splits. It describes the data in terms of queries for evaluation, not as a split for model training.
Hardware Specification Yes Experiments are run on a 3.4 GHz machine with 16 GB of memory.
Software Dependencies No The paper mentions using 'PLP-to-WMC conversion of Fierens et al. (2011)', 'compilation to SDDs (Darwiche 2011)', and the 'WPMS solver...WPM1 (Ans otegui, Bonet, and Levy 2009)'. However, it does not provide specific version numbers for these software components.
Experiment Setup Yes We select 150 pages, and we use the 100 most frequent words only. Probabilities are chosen randomly from [0.1, 0.4]. [...] we give each algorithm a time budget of five minutes per query. [...] We further consider three different thresholds (0.05, 0.01, 0.0001) on the quality of approximation as measured by the maximal error...