Optimal Bipartite Network Clustering

Authors: Zhixin Zhou, Arash A. Amini

JMLR 2020 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show the effectiveness of this general class by numerical simulations. ... We provide some simulation results to corroborate the theory. We generate from the SBM of Section 2.1 with the following connectivity matrix... Figure 3 shows the results for α = .75 (a regime where no exact recovery is possible) and C = 1. Both the soft and hard versions of Algorithm 1 are initialized with the spectral clustering and both significantly improve over it.
Researcher Affiliation Academia Zhixin Zhou EMAIL Department of Statistics University of California Los Angeles, CA 90095-1554, USA; Arash A. Amini EMAIL Department of Statistics University of California Los Angeles, CA 90095-1554, USA
Pseudocode Yes Algorithm 1 Pseudo-likelihood biclustering, meta algorithm; Algorithm 2 Simplified pseudo-likelihood clustering; Algorithm 3 Provable (parallelizable) version; Algorithm 4 SC-RRE
Open Source Code No The paper does not provide any explicit statements about the release of source code for the described methodology, nor does it include links to a code repository.
Open Datasets No We generate from the SBM of Section 2.1 with the following connectivity matrix... We let n = Kn0 and m = Ln0, and we vary n0. All clusters (both row and column) will have the same number of nodes n0.
Dataset Splits No The paper describes generating synthetic data based on a stochastic block model for simulations, but it does not mention or define any specific training, testing, or validation splits for a dataset.
Hardware Specification No The paper does not provide specific details about the hardware used to run the simulations or experiments.
Software Dependencies No The paper describes the algorithms and theoretical results, and discusses numerical simulations, but does not specify any software dependencies or their version numbers (e.g., programming languages, libraries, or frameworks).
Experiment Setup Yes We generate from the SBM of Section 2.1 with the following connectivity matrix... We let n = Kn0 and m = Ln0, and we vary n0. All clusters (both row and column) will have the same number of nodes n0. By changing α, we can study different regimes of sparsity. ... Figure 3 shows the results for α = .75 (a regime where no exact recovery is possible) and C = 1. ... We have considered four algorithms: (1) Spectral: the spectral clustering of Algorithm 4. (2) Soft: Algorithm 1 with flat prior, no inner loop and no conversion to hard labels. (3) Hard: Algorithm 1 with flat prior, no inner loop and conversion to hard labels after each label computation. (4) Oracle: The oracle classifier discussed in Section 2.2 and Remark 2: Assuming the knowledge of z and Λ, we obtain by by the likelihood ratio classifier, and similarly obtain bz, assuming the knowledge of y and Γ.