Discrete and Continuous Difference of Submodular Minimization

Authors: George Orfanides, Tim Hoheisel, Marwa El Halabi

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate that our method outperforms baselines in integer compressive sensing and integer least squares.
Researcher Affiliation Collaboration 1Department of Mathematics and Statistics, Mc Gill University, Montreal 2Samsung AI Lab, Montreal.
Pseudocode Yes Algorithm 1 DCA with local search
Open Source Code Yes The code is available at https://github.com/SamsungSAILMontreal/cont-diffsubmin.
Open Datasets No We sample x uniformly from X = {−1, 0, 2, 3}n with n = 100, draw the entries of A i.i.d from N(0, 1), and vary m from n to 2n. ... We set X = {−1, 0, 1}n, n = 256, s = 26 = 0.1n, and draw A i.i.d from N(0, 1/m), with m varied from 26 to n.
Dataset Splits No The paper describes generating synthetic data for integer least squares and integer compressed sensing problems, but does not refer to pre-existing datasets nor specify any training/test/validation splits for such datasets. It defines parameters for data generation and evaluation.
Hardware Specification No The paper mentions that 'All methods were implemented in MATLAB' but does not specify any particular hardware (e.g., CPU, GPU models) used for running the experiments.
Software Dependencies Yes We obtain an optimal solution x∗ using Gurobi 10.0.1 (Gurobi Optimization, LLC, 2024).
Experiment Setup Yes We sample x uniformly from X = {−1, 0, 2, 3}n with n = 100, draw the entries of A i.i.d from N(0, 1), and vary m from n to 2n. The noise variance σ2 is set to achieve a target signal-to-noise ratio (SNRdB) of 20 dB. ... In DCA-LS, we use a maximum of T = 50 outer iterations and set ϵ = 10−5.