Greed is Still Good: Maximizing Monotone Submodular+Supermodular (BP) Functions

Authors: Wenruo Bai, Jeff Bilmes

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Computational Experiments We empirically test our guarantees for BP maximization subject to a cardinality constraint on contrived functions using GREEDMAX and Semi Grad.
Researcher Affiliation Academia 1Department of Electrical Engineering, University of Washington, Seattle, USA 2Department of Computer Science and Engineering, University of Washington, Seattle, USA.
Pseudocode Yes Algorithm 1: GREEDMAX for BP maximization
Open Source Code No The paper does not contain an explicit statement or link indicating the release of source code for the described methodology.
Open Datasets No For the first experiment, we let |V | = 20 set the cardinality constraint to k = 10, and partition the ground set into |V1| = |V2| = k, V1 V2 = V where V1 = {v1, v2, . . . , vk}. Let wi = 1 α h 1 α k i+1i for i = 1, 2, . . . , k. Then we define the submodular and supermodular functions as follows, f(X) = h k α|X V2| {i:vi X} wi + |X V2| g(X) = |X| β min(1 + |X V1|, |X|, k) + ϵ max(|X|, |X| + β 1 β (|X V2| k + 1)) and h(X) = λf(X)+(1 λ)g(X) for 0 α, β, λ 1 and ϵ = 1 10 5.
Dataset Splits No No explicit mention of specific dataset split percentages, sample counts, or references to predefined splits for training, validation, or test sets.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for computational experiments.
Software Dependencies No The paper does not provide specific details about ancillary software, including names and version numbers.
Experiment Setup Yes For the first experiment, we let |V | = 20 set the cardinality constraint to k = 10, and partition the ground set into |V1| = |V2| = k, V1 V2 = V where V1 = {v1, v2, . . . , vk}. Let wi = 1 α h 1 α k i+1i for i = 1, 2, . . . , k. Then we define the submodular and supermodular functions as follows, f(X) = h k α|X V2| {i:vi X} wi + |X V2| g(X) = |X| β min(1 + |X V1|, |X|, k) + ϵ max(|X|, |X| + β 1 β (|X V2| k + 1)) and h(X) = λf(X)+(1 λ)g(X) for 0 α, β, λ 1 and ϵ = 1 10 5. Immediately, we notice that κf = α and κg = β. In particular, we choose α, β, λ = 0, 0.01, 0.02, . . . , 1 and for all cases, we normalize h(X) using either exhaustive search so that OPT = h(X ) = 1.