AI Paper Atlas

Research map / Learning theory and algorithms / Matrix and tensor methods

Unique Sharp Local Minimum in L1-minimization Complete Dictionary Learning

Yu Wang, Siqi Wu, Bin Yu

JMLR 2020 · Learning theory and algorithms › Matrix and tensor methods · approach: Sparse regression with guarantees

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
increase perturbation parameter rho beyond 0.35 in the sharpness test (ablation)Algorithm 1 (perturbation test) · Sharp dictionary, K=20, s=10, n=1600distance r: ~10^{-12} -> ~10^{-3}Hurt
use rho between 0.05 and 0.3 in the sharpness test (ablation)Algorithm 1 (perturbation test) · Sharp vs non-sharp dictionaries, K=20distinguishing sharp from non-sharp: around 1.0 (non-sharp) vs ~10^{-12} (sharp)Improved
increase dimension K from 12 to 20 in sharpness identification (ablation)Algorithm 1 (perturbation test) · Sparse Gaussian s=5, mu=0.5sample size for 50% chance: 253, 316, 375 for K = 12, 16, 20Hurt
use the DL-BCD algorithm with tau=.5 instead of K-SVD, SPAMS, ER-SpUD, EM-BiG-AMPcomplete dictionary learning · Recovery rate, n=100K, SNR 100, K=2..20recovery rateImproved
replace Gaussian with Laplace coefficientslocal identifiability condition · Phase transition, K=10 and 20phase transition curveImproved
use reweighting by truncating coefficients above threshold tau (ablation)DL-BCD · high SNR dictionary learningnumerical performanceImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy