What Are the Rules? Discovering Constraints from Data

Authors: Boris Wiegand, Dietrich Klakow, Jilles Vreeken

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on constraint programming and AI planning benchmark data show URPILS not only finds more accurate and succinct constraints, but also is more robust to noise, and has lower sample complexity than the state of the art. In summary, our main contributions are as follows. We (a) formalize the problem of learning constraints from exemplary solutions in terms of the MDL principle, (b) propose an efficient heuristic to discover constraints for both constraint programming and AI planning, (c) provide an extensive empirical evaluation, (d) make code, data and additional details publicly available in the supplementary materials.
Researcher Affiliation Collaboration Boris Wiegand1,2, Dietrich Klakow2, Jilles Vreeken3 1 SHS Stahl-Holding-Saar, Dillingen, Germany 2 Saarland University, Saarbr ucken, Germany 3 CISPA Helmholtz Center for Information Security, Germany
Pseudocode Yes Algorithm 1: URPILS; Algorithm 2: FILTER
Open Source Code Yes To ensure reproducibility, we make code and data publicly available in the extra materials.1 1https://eda.rg.cispa.io/prj/urpils
Open Datasets Yes We generate valid assignments for all datasets and split them into training and test set. To ensure reproducibility, we make code and data publicly available in the extra materials.1 1https://eda.rg.cispa.io/prj/urpils
Dataset Splits No The paper mentions splitting data into training and test sets but does not explicitly specify a separate validation split or its details.
Hardware Specification Yes We conducted all our experiments on a PC with Windows 10, an Intel i7-6700 CPU and 32 GB of memory.
Software Dependencies No The paper mentions Windows 10 as the operating system but does not provide specific version numbers for any other software dependencies, libraries, or frameworks used in the experiments.
Experiment Setup Yes While COUNTOR and COUNTCP have no hyperparameters, we must generate candidate constraints for MINEACQ and set parameters τ and ρ to control the acceptance threshold of its permutation test for candidate selection. By a manual hyperparameter search, we find τ = 10 and ρ = 0.001 lead to the best results.