Fast greedy algorithms for dictionary selection with generalized sparsity constraints

Authors: Kaito Fujii, Tasuku Soma

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

Reproducibility Variable Result LLM Response
Research Type Experimental Using numerical experiments, we show that our algorithm outperforms the known methods for dictionary selection, achieving competitive performances with dictionary learning algorithms in a smaller running time.
Researcher Affiliation Academia Kaito Fujii Graduate School of Information Sciences and Technology The University of Tokyo kaito_fujii@mist.i.u-tokyo.ac.jp Tasuku Soma Graduate School of Information Sciences and Technology The University of Tokyo tasuku_soma@mist.i.u-tokyo.ac.jp
Pseudocode Yes Algorithm 1 Replacement Greedy & Replacement OMP
Open Source Code No The paper states 'All the algorithms are implemented in Python 3.6.' and that they implemented their proposed and existing methods, but it does not provide an explicit statement about releasing their code or a link to a code repository.
Open Datasets Yes The second one is a dataset of real-world images extracted from PASCAL VOC2006 image datasets [15].
Dataset Splits No The paper mentions 'We make datasets for training and test in the same way, and use the training dataset for obtaining a dictionary and the test dataset for measuring the quality of the output dictionary.' but does not specify exact percentages, sample counts, or the methodology for these splits (e.g., random, stratified, or predefined splits with citations).
Hardware Specification Yes We conduct the experiments in a machine with Intel Xeon E3-1225 V2 (3.20 GHz and 4 cores) and 16 GB RAM.
Software Dependencies No The paper states 'All the algorithms are implemented in Python 3.6.' but does not list any specific libraries or solvers with their version numbers beyond the programming language itself.
Experiment Setup Yes The parameter of sparsity constraints is set to s = 5. The parameters of constraints are set to st = 8 for all t [T] and s = 5T. The dimension is set to d = 64, which corresponds to images of size 8 x 8 pixels. The size of the ground set is n = 256. k = 20 and s = 5 for the online setting. KSVD [2], which is set to stop when the change of the objective value becomes no more than 10^-6 or 200 iterations are finished.