Research map / Learning theory and algorithms / Matrix and tensor methods
Complete Dictionary Learning via L4-Norm Maximization over the Orthogonal Group
Yuexiang Zhai, Zitong Yang, Zhenyu Liao, John Wright, Yi Ma
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 changed | Where | Measured | Effect |
|---|---|---|---|
| increase PGA step size alpha from 1 to +infinity (MSP) on O(200) (ablation) | PGA l4 maximization · O(200;R) | iterations to global maximizer: 70 -> 9 | Improved |
| increase PGA step size alpha from 1 to +infinity (MSP) on O(100) (ablation) | PGA l4 maximization · O(100;R) | iterations to global maximizer: 63 -> 9 | Improved |
| increase PGA step size alpha from 1 to +infinity (MSP) on O(5) (ablation) | PGA l4 maximization · O(5;R) | iterations to global maximizer: 13 -> 4 | Improved |
| increase step size from alpha=100 to +infinity on O(100) (ablation) | PGA l4 maximization · O(100;R) | iterations to global maximizer: 9 -> 9 | No clear effect |
| use higher-order l2k norm (larger k) in deterministic orthogonal-group maximization (ablation) | MSP · deterministic case | convergence speed to signed permutation | Improved |
| use higher-order l2k norm (larger k) in dictionary learning (ablation) | MSP · n=10, varying k and p | samples needed for same estimation error | Hurt |
| MSP vs KSVD, SPAMS and subgradient method for dictionary learning | MSP · theta=0.3, varying n and p | running time: 374.2 seconds for 400x400 dictionary from 160,000 samples | Improved |
| remove ground-truth dependence from stopping criterion (MSP vs subgradient method) | MSP · theta=0.3, varying n and p | l4 norm of A D_o over n | Mixed |
| raise sparsity level theta toward 0.6 with large enough p (ablation) | MSP · n=50, theta 0 to 1 | dictionary recovery: theta ~ 0.6 | No clear effect |
Closest papers by meaning
- Understanding l4-based Dictionary Learning: Interpretation, Stability, and Robustness (ICLR 2020 · similarity 0.93)
- Alternating minimization for dictionary learning with random initialization (NeurIPS 2017 · similarity 0.87)
- Dictionary Learning Based on Sparse Distribution Tomography (ICML 2017 · similarity 0.87)
- Unique Sharp Local Minimum in L1-minimization Complete Dictionary Learning (JMLR 2020 · similarity 0.87)
- Hiding Data Helps: On the Benefits of Masking for Sparse Coding (ICML 2023 · similarity 0.87)
- Dictionary Learning for Sparse Coding: Algorithms and Convergence Analysis (TPAMI 2016 · similarity 0.87)
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.
