Efficient Near-Optimal Testing of Community Changes in Balanced Stochastic Block Models

Authors: Aditya Gangrade, Praveen Venkatesh, Bobak Nazer, Venkatesh Saligrama

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate these phenomena numerically on SBMs and on real-world datasets as well as Markov Random Fields where we only observe node data rather than the existence of links. We complement the above theoretical study by three experiments: the first implements the above tests on synthetic SBMs, and the second on the political blogs dataset a popular real world dataset for community detection [AG05]. Both of these experiments show excellent agreement with the theoretical predictions.
Researcher Affiliation Academia Aditya Gangrade Boston University gangrade@bu.edu Praveen Venkatesh Carnegie Mellon University vpraveen@cmu.edu Bobak Nazer Boston University bobak@bu.edu Venkatesh Saligrama Boston University srv@bu.edu
Pseudocode Yes Algorithm 1: Two Sample Tester(G, H, δ)
Open Source Code No The paper does not contain any explicit statement about releasing source code or provide links to a code repository for the methodology described.
Open Datasets Yes Next, we demonstrate that our tests perform similarly for a real dataset, specifically the Political Blogs dataset [AG05].
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits (e.g., exact percentages or sample counts).
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory) used to run its experiments.
Software Dependencies No The paper mentions using Python libraries such as SciPy, NumPy, Matplotlib, and Scikit-learn, but it does not specify their version numbers.
Experiment Setup Yes Recovery is performed by regularised spectral clustering, for which a detailed description is given in Appendix C.1. In Appendix C.1, it states: 'Spectral clustering requires a choice of k. In all experiments we set k = 2. It also requires a normalisation for the adjacency matrix. For the SBM experiments, we use the normalised adjacency matrix A = D−1/2AD−1/2. The regularised spectral clustering for sparse graphs is described in [JY16]. It involves adding µnI to the adjacency matrix prior to computing the eigenvalues. We set µn = p log n/(n(a + b)).'