A General Efficient Hyperparameter-Free Algorithm for Convolutional Sparse Learning

Authors: Zheng Xu, Junzhou Huang

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments confirm the superior performance of our general algorithm in various convolutional sparse models, even better than some application-specialistic algorithms.
Researcher Affiliation Academia Zheng Xu, Junzhou Huang Department of Computer Science and Engineering The University of Texas at Arlington
Pseudocode Yes Algorithm 1 Primal-Dual Algorithm for Convolutional Sparsity Let ζ = m i=1 ki 2 1, L is a Lipschitz smooth constant of f. Choose β(0) X, μ(0) = (μ(0) 1 , μ(0) 2 , . . . , μ(0) m ) U = U1 U2 Um. Iterate: for t = 0, 1, 2, . . . Update primal variable: β(t+1) = prox g L+ζ β(t) 1/(L + ζ)[ f(β(t)) + CH [m]μ(t)] . Update dual variable: μ(t+1) = ΠB (μ(t) + (1/ζ)C[m](2β(t+1) β(t))).
Open Source Code No The paper does not provide an explicit statement or link for open-source code for the described methodology.
Open Datasets No In this experiment, we hereby set X Rn p as a normal Gaussian matrix, and the smallest Lipschitz smooth constant of f(β) has been estimated as O(( n + p)2) according to (Rudelson and Vershynin 2010). β Rp is randomly generated as ground truth. ... We compare our method with SLEP (Liu, Ji, and Ye 2009)3, Chambolle & Pock (hereafter C & P) (Chambolle and Pock 2014), CVX (Grant and Boyd 2014) and f GFL (Xin et al. 2014). ... Joint Total Variation and Nuclear Norm Regularization ... It is written as the following optimization problems, min β 1 2 Xβ y 2 F + λ1 β + λ2 β 1. (14) Here, β has two spatial dimensions (i.e., a matrix), with X serving as a linear bounded subsampling operator.
Dataset Splits No The paper mentions numerical experiments but does not explicitly state the training, validation, or test dataset splits.
Hardware Specification Yes All the experiments in this section are conducted on a desktop computer with Intel Core i7-4770 CPU and 16 gigabyte RAM.
Software Dependencies No All methods are evaluated in MATLAB 2013b, Windows 7 Enterprise.
Experiment Setup Yes In this experiment, we hereby set X Rn p as a normal Gaussian matrix, and the smallest Lipschitz smooth constant of f(β) has been estimated as O(( n + p)2) according to (Rudelson and Vershynin 2010). β Rp is randomly generated as ground truth. We use the exact same parameter setting as in (Liu, Yuan, and Ye 2010), i.e., λ1 = 0.001, λ2 = 0.01. ... The parameter setting is λ1 = 100, λ2 = 1, following the same setting in (Huang et al. 2011).