A Simple and Strongly-Local Flow-Based Method for Cut Improvement

Authors: Nate Veldt, David Gleich, Michael Mahoney

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the power of Simple Local by solving problems on a 467 million edge graph based on an MRI scan. In this section we present experimental results for Simple Local on two graphs. We begin with an example on a small collaboration network to illustrate the effect of the locality parameter δ. We then turn our attention to graphs from MRI scans to demonstrate Simple Local s ability to solve problems on extremely large graphs.
Researcher Affiliation Academia Nate Veldt lveldt@purdue.edu Mathematics Department, Purdue University, West Lafayette, IN 47906 David F. Gleich dgleich@purdue.edu Computer Science, Purdue University, West Lafayette, IN 47906 Michael W. Mahoney mmahoney@stat.berkeley.edu International Computer Science Institute and Dept. of Statistics, University of California at Berkeley, Berkeley, CA 94720
Pseudocode Yes An outline for 3Stage Flow is given in Algorithm 1. Algorithm 2 Simple Local
Open Source Code No The paper states 'Our implementation of Simple Local and 3Stage Flow are in Matlab, using Gurobi to solve the max-flow problems,' but it does not provide any explicit statement about open-sourcing the code or a link to a repository for the methodology.
Open Datasets Yes We obtained a labeled MRI scan from the MICCAI-2012 challenge with 256 287 256 (~18 million) voxels (Marcus et al., 2007). The MRI scans originated with the OASIS project and labeled data was provided by Neuromorphometrics, Inc. neuromorphometrics.com under an academic subscription.
Dataset Splits No The paper mentions using '75 randomly chosen seed voxels' for the MRI scan experiment, but it does not specify any training, validation, or test dataset splits or a cross-validation setup for reproducibility.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as CPU/GPU models, processor types, or memory amounts.
Software Dependencies No The paper states 'Our implementation of Simple Local and 3Stage Flow are in Matlab, using Gurobi to solve the max-flow problems,' but it does not specify version numbers for Matlab or Gurobi, which are necessary for reproducible software dependencies.
Experiment Setup No The paper discusses the locality parameter δ and mentions using '75 randomly chosen seed voxels.' However, it defers detailed computational and parameter choices for the MRI scans to the supplementary material ('See the supplement for the details of the computations and parameter choices.'), thus not providing them in the main text.