Learning Erdos-Renyi Random Graphs via Edge Detecting Queries

Authors: Zihan Li, Matthias Fresacher, Jonathan Scarlett

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

Reproducibility Variable Result LLM Response
Research Type Experimental We complement our theoretical findings with numerical experiments comparing COMP, DD, SSS, and a linear programming (LP) relaxation of SSS (analogous to [34]).5 Figure 2 shows the success probability as a function of the number of tests in two cases: (i) n = 50 and k = 5; (ii) n = 200 and k = 200. In each case, we set ν = 1 and compute the error probability averaged over 2000 trials.
Researcher Affiliation Academia Zihan Li National University of Singapore lizihan@u.nus.edu Matthias Fresacher University of Adelaide matthias.fresacher@adelaide.edu.au Jonathan Scarlett National University of Singapore scarlett@comp.nus.edu.sg
Pseudocode Yes Algorithm 1 Combinatorial Orthogonal Matching Pursuit (COMP) Algorithm 2 Definite Defectives (DD) Algorithm 3 Smallest Satisfying Set (SSS) Algorithm 4 Group Testing Quick and Efficient (GROTESQUE) Informal Outline
Open Source Code Yes 5The code is available at https://github.com/scarlett-nus/er_edge_det.
Open Datasets No The paper uses the Erd os-Rényi random graph model to generate graphs for its experiments, rather than relying on an external, pre-existing publicly available dataset. Therefore, no specific information about public dataset access is provided.
Dataset Splits No The numerical experiments are based on simulated data generated from the Erd os-Rényi random graph model with specified parameters (e.g., n=50, k=5). The concept of training/validation/test splits, as typically applied to fixed datasets, is not applicable or mentioned in this simulation context.
Hardware Specification No The paper mentions numerical experiments but does not provide any specific details about the hardware used to run them.
Software Dependencies No The paper does not explicitly provide specific version numbers for any software dependencies (e.g., programming languages, libraries, or solvers) used for the experiments.
Experiment Setup Yes In each case, we set ν = 1 and compute the error probability averaged over 2000 trials.