Adaptive Shrinkage Estimation for Streaming Graphs

Authors: Nesreen Ahmed, Nick Duffield

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments on large networks show that our approach is superior to baseline methods.
Researcher Affiliation Collaboration Nesreen K. Ahmed Intel Labs Santa Clara, CA 95054 nesreen.k.ahmed@intel.com Nick Duffield Texas A&M University College Station, TX 77843 duffieldng@tamu.edu
Pseudocode Yes Algorithm 1 Adaptive Priority Sampling (APS)
Open Source Code No The paper does not provide a statement or link for the open-sourcing of its code.
Open Datasets Yes Experimental Setup. We test on graphs from different domains and with different characteristics; see [40] for data downloads. Table 1 provides a summary of dataset characteristics, where |V | is the number of vertcies, |K| is the number of edges, T is the number of triangles, and Tmax is the maximum triangle count per edge.
Dataset Splits No The paper discusses "sample fractions" for the sampling methods but does not define training, validation, and test splits for a machine learning model, nor does it specify cross-validation.
Hardware Specification Yes All experiments were performed using a server with two Intel Xeon E5-2687W 3.1GHz CPUs, 256GB of memory.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes Experimental Setup. We test on graphs from different domains and with different characteristics; see [40] for data downloads. ... We repeat the experiment ten different times with sample fractions f = {0.10, 0.20, 0.40, 0.50}. ... Our experimental setup is summarized as follows: For each sample fraction, we use Algorithm 1 to collect a sample b K, from edge stream K. The experiments in this section use triangles as an example of the motif pattern M. ... We compare the results of Algorithm 1 with uniform sampling (i.e., reservoir sampling [53]) using the Horvitz-Thompson estimator, and we also compare with Triest sampling [48].