Continuous DR-submodular Maximization: Structure and Algorithms

Authors: An Bian, Kfir Levy, Andreas Krause, Joachim M. Buhmann

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our theoretical findings are validated on synthetic and real-world problem instances.
Researcher Affiliation Academia An Bian ETH Zurich ybian@inf.ethz.ch Kfir Y. Levy ETH Zurich yehuda.levy@inf.ethz.ch Andreas Krause ETH Zurich krausea@ethz.ch Joachim M. Buhmann ETH Zurich jbuhmann@inf.ethz.ch
Pseudocode Yes Algorithm 1: TWO-PHASE FRANK-WOLFE for non-monotone DR-submodular maximization
Open Source Code Yes All experiments were implemented using MATLAB. Source code can be found at: https://github.com/bianan/non-monotone-dr-submodular.
Open Datasets Yes In our experiments, we used a similar setting to the one in [20]. We experimented on the 2012 US Republican debates data, which consists of 8 candidates: Bachman, Gingrich, Huntsman, Paul, Perry, Romney and Santorum.
Dataset Splits No The paper mentions using synthetic and real-world data but does not provide explicit details on train/validation/test splits, such as percentages, sample counts, or references to predefined splits.
Hardware Specification No No specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments are mentioned in the paper.
Software Dependencies No The paper states 'All experiments were implemented using MATLAB.' and refers to 'QUADPROGIP2 [39]' and 'IBM CPLEX optimization studio' as subroutines. However, specific version numbers for MATLAB or CPLEX are not provided, preventing full reproducibility of the software environment.
Experiment Setup Yes We run all the algorithms for 100 iterations. For the subroutine (Algorithm 3) of TWO-PHASE FRANK-WOLFE, we set 1 = 2 = 10 6, K1 = K2 = 100. ... In order to make f non-monotone, we set h = 0.2 H> u. To make sure that f is non-negative, we first of all solve the problem minx2P 1 2x>Hx + h>x using QUADPROGIP, let the solution to be ˆx, then set c = f(ˆx) + 0.1 |f(ˆx)|.