Local Linear Convergence of Forward--Backward under Partial Smoothness

Authors: Jingwei Liang, Jalal Fadili, Gabriel Peyré

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Several experimental results on some of the problems discussed above are provided to support our theoretical findings. 4 Numerical experiments In this section, we describe some examples to demonstrate the applicability of our results. More precisely, we consider solving min x Rn 1 2||y Ax||2 + λJ(x) (4.1)
Researcher Affiliation Academia Jingwei Liang and Jalal M. Fadili GREYC, CNRS-ENSICAEN-Univ. Caen {Jingwei.Liang,Jalal.Fadili}@greyc.ensicaen.fr Gabriel Peyré CEREMADE, CNRS-Univ. Paris-Dauphine Gabriel.Peyre@ceremade.dauphine.fr
Pseudocode No The paper describes the Forward Backward (FB) splitting algorithm using a mathematical update rule (1.2) but does not provide it in a structured pseudocode or algorithm block format.
Open Source Code No The paper does not provide any concrete access to source code for the methodology described.
Open Datasets No The paper describes generating synthetic data for its experiments (e.g., 'A Rm n is generated uniformly at random from the Gaussian ensemble', 'x0 is 8-sparse', 'rank(x0) = 5') rather than using publicly available datasets with concrete access information.
Dataset Splits No The paper describes the experimental settings for various problems but does not provide specific dataset split information (e.g., percentages or counts for training, validation, or test sets).
Hardware Specification No The paper does not provide any specific hardware details used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes The tested experimental settings are (a) ℓ1-norm m = 48 and n = 128, x0 is 8-sparse; (b) Total Variation m = 48 and n = 128, (DDIFx0) is 8-sparse; (c) ℓ∞-norm m = 123 and n = 128, x0 has 10 saturating entries; (d) ℓ1 ℓ2-norm m = 48 and n = 128, x0 has 2 non-zero blocks of size 4; (e) Nuclear norm m = 1425 and n = 2500, x0 R50 50 and rank(x0) = 5. The number of measurements is chosen sufficiently large, δ small enough and λ of the order of δ so that [27, Theorem 1] applies, yielding that the minimizer of (4.1) is unique and verifies the non-degeneracy and restricted strong convexity assumptions (3.1)-(3.2).