Learning-Augmented Dynamic Submodular Maximization

Authors: Arpit Agarwal, Eric Balkanski

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 7 Experiments
Researcher Affiliation Academia Arpit Agarwal Indian Institute of Technology Bombay aarpit@iitb.ac.in Eric Balkanski Columbia University eb3224@columbia.edu
Pseudocode Yes Algorithm 1 The Algorithmic Framework; Algorithm 2 WARMUP-UPDATESOL; Algorithm 3 UPDATESOLMAIN; Algorithm 4 PRECOMPUTATIONSMAIN; Algorithm 5 ROBUST1FROMDYNAMIC; Algorithm 6 PRECOMPUTATIONSFULL; Algorithm 7 UPDATESOLFULL
Open Source Code No We are not submitting the code because one of the libraries we extensively use/modify requires several conditions for distributing derivatives of their library. We did not have time to satisfy all these conditions before the deadline. However, we have carefully read these conditions and will definitely be able to meet these conditions before the potential camera-ready deadline, at which point we would release our code.
Open Datasets Yes We perform experiments on a subset of the Enron dataset from the SNAP Large Networks Data Collection [21].
Dataset Splits No The paper describes a sliding window protocol for generating a dynamic stream of insertions and deletions and sets parameters like window size, but it does not specify explicit training, validation, or test dataset splits in terms of percentages or sample counts.
Hardware Specification No we only compute small scale experiments which can be run within a few minutes on a CPU.
Software Dependencies No We implemented our algorithm and OFFLINEGREEDY in C++, and used the C++ implementation of DYNAMIC that is provided by [20]. No specific version numbers for C++ or related libraries are mentioned.
Experiment Setup Yes We set ϵ = 0.2 for all algorithms.