Engineering an Efficient Approximate DNF-Counter

Authors: Mate Soos, Divesh Aggarwal, Sourav Chakraborty, Kuldeep S. Meel, Maciej Obremski

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our approach through extensive experiments and show that it provides an affirmative answer to the challenge of efficiently computing #DNF. 5 Empirical Evaluation
Researcher Affiliation Academia Mate Soos1 , Divesh Aggarwal1,3 , Sourav Chakraborty 2 , Kuldeep S. Meel1 , Maciej Obremski3 1National University of Singapore, Singapore 2Indian Statistical Institute, Kolkata 3The Centre for Quantum Technologies (CQT), Singapore
Pseudocode Yes Algorithm 1 MVC, Algorithm 2 pepin(F, ε, δ), Algorithm 3 Compute Num Samples(t, p), Algorithm 4 Generate Samples(Ni, Fi), Algorithm 5 Compute Num Samples Binom(t, p)
Open Source Code Yes The corresponding open-source tool is available at https:// github.com/meelgroup/pepin
Open Datasets No The paper states, 'we use the same benchmark generation tool with the parameters specified by the authors,' implying the datasets were generated rather than being a publicly available, pre-existing dataset with a direct link or citation provided for access.
Dataset Splits No The paper describes running experiments on a 'benchmark suite of 900 instances' generated by a 'benchmark generation tool' but does not specify any training, validation, or test dataset splits in the conventional sense for model training.
Hardware Specification Yes All our experiments were conducted on a highperformance computer cluster, each node consisting of 2x E52690v3 CPUs with 2x12 real cores and 96GB of RAM, i.e., 4GB limit per run.
Software Dependencies No The paper mentions making 'extensive use of the GNU Bignum library' but does not provide a specific version number for this software dependency within the text.
Experiment Setup Yes Furthermore, in line with prior work, we set ε to 0.8 and δ to 0.36. ... The timeout was set to be 500 seconds for all runs.