Horizontally Scalable Submodular Maximization

Authors: Mario Lucic, Olivier Bachem, Morteza Zadimoghaddam, Andreas Krause

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

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluate the proposed algorithm on a variety of data sets and demonstrate that it achieves performance competitive with the centralized greedy solution.
Researcher Affiliation Collaboration 1Department of Computer Science, ETH Zurich, Switzerland 2Google Research, New York
Pseudocode Yes Pseudo-code is provided in Algorithm 1 and one round is illustrated in Figure 1.
Open Source Code No The paper does not include an unambiguous statement or a direct link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes CSN. The Community Seismic Network uses smart phones with accelerometers as inexpensive seismometers for earthquake detection. In Faulkner et al. (2011), 7 GB of acceleration data was recorded... TINY IMAGES. In our experiments we used two subsets of the Tiny Images data set consisting of 32 32 RGB images... (Torralba et al., 2008). PARKINSONS. The data set consists of 5875 biomedical voice measurements... (Tsanas et al., 2010).
Dataset Splits No The paper mentions subsampling (e.g., 'We select a fixed random subsample of 10 000 elements for evaluation on each machine') but does not specify explicit training, validation, or test dataset splits or percentages for reproducing the experiments.
Hardware Specification No The paper mentions 'machines of fixed capacity' conceptually within its framework description, but it does not provide specific hardware details such as GPU or CPU models, processor types, or memory specifications used for running experiments.
Software Dependencies No The paper mentions specific algorithms like GREEDY and STOCHASTIC GREEDY, but it does not list any specific software libraries or dependencies with their version numbers that would be needed to replicate the experiments.
Experiment Setup Yes We consider three baseline methods... We use the lazy variant of the GREEDY algorithm... For each algorithm we report the ratio between the obtained function value and the one obtained by the centralized GREEDY averaged over 10 trials... The capacity is set to a small percentage of the ground set size (0.05% and 0.1%). Furthermore, we consider two instances of STOCHASTIC GREEDY, one with epsilon = 0.5 and the other with epsilon = 0.2... We perform several experiments optimizing the active set selection objective with a Gaussian kernel (h = 0.5 and sigma = 1).