Difference of submodular minimization via DC programming
Authors: Marwa El Halabi, George Orfanides, Tim Hoheisel
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our numerical results show that our proposed algorithms outperform existing baselines on two applications: speech corpus selection and feature selection. |
| Researcher Affiliation | Collaboration | 1Samsung SAIT AI Lab, Montreal 2Department of Mathematics and Statistics, Mc Gill University. |
| Pseudocode | Yes | Algorithm 1 Approximate DCA |
| Open Source Code | Yes | The code is available at https://github.com/Samsung SAILMontreal/ difference-submodular-min.git. |
| Open Datasets | Yes | We use the same dataset used by (Bach, 2013, Section 12.1), with d = |V | = 800 utterances and 1105 words. ... We use the Mushroom data set from (Dua & Graff, 2017), which has 8124 instances with 22 categorical attributes, which we convert to d = 118 binary features. |
| Dataset Splits | No | The paper states, "We randomly select 70% of the data as training data for the feature selection," but it does not specify percentages or details for validation or testing splits, nor does it explicitly mention a validation set. |
| Hardware Specification | No | The paper does not explicitly specify any hardware details such as GPU/CPU models, memory, or cloud computing instances used for running experiments. |
| Software Dependencies | No | The paper states, "We use the implementation of MNP from the Matlab code provided in (Bach, 2013, Section 12.1) and implement the rest of the methods in Matlab." While Matlab is mentioned, no specific version numbers for Matlab or any associated libraries are provided. |
| Experiment Setup | Yes | We choose λ = 1, the non-negative weights mi randomly, and partition V into r = 10 groups of consecutive indices. ... We randomly select 70% of the data as training data for the feature selection, and set λ = 10 4. ... We use f(xk) f(xk+1) 10 6 as a stopping criterion in our methods... We set the parameter q in ADCA and ADCAR to 5... We summarize the stopping criteria used in all methods and their subsolvers in Table 1. |