Research map / Optimization / Implicit bias and generalization
Sparse GCA and Thresholded Gradient Descent
Sheng Gao, Zongming Ma
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 |
|---|---|---|---|
| add thresholded gradient descent refinement after Fantope initialization | Generalized Fantope initial estimator · Sparse GCA 3 datasets, r=1 | squared matrix distance: 0.1319 -> 0.0015 | Improved |
| add thresholded gradient descent refinement after Fantope initialization | Generalized Fantope initial estimator · Sparse GCA 3 datasets, r=5 | squared matrix distance: 0.1969 -> 0.0171 | Improved |
| use TGD instead of Rifle for refinement | Rifle · Sparse GCA r=1, n=500 | squared matrix distance: 0.0006 -> 0.0003 | Improved |
| use TGD instead of Rifle for refinement | Rifle · Sparse GCA r=1, n=1000 | squared matrix distance: 0.0003 -> 0.0002 | Improved |
| use TGD instead of CoLaR for refinement | CoLaR · Sparse CCA Toeplitz (300,300,200) | prediction loss: 0.0511 -> 0.0118 | Improved |
| use TGD instead of CoLaR for refinement | CoLaR · Sparse CCA Toeplitz (600,600,500) | prediction loss: 0.0176 -> 0.0213 | Hurt |
| use TGD instead of CoLaR for refinement | CoLaR · Sparse CCA SparseInv (300,300,500) | prediction loss: 0.0242 -> 0.0042 | Improved |
| use TGD instead of CoLaR for refinement | CoLaR · Sparse CCA Identity (600,600,500) | prediction loss: 0.0166 -> 0.0082 | Improved |
| vary penalty lambda from 0.0001 to 1 (ablation) | TGD · Sparse GCA r=3 | squared matrix distance: 0.0121 -> 0.0099 | No clear effect |
| vary penalty lambda from 0.0001 to 1 (ablation) | TGD · Sparse GCA r=4 | squared matrix distance: 0.0145 -> 0.0119 | No clear effect |
| choose sparsity level s' by 5-fold cross-validation (ablation) | TGD · Sparse GCA r=3 (s=15) | selected s': 25 | No clear effect |
| input latent dimension r=2 when true dimension is 3 (misspecification) (ablation) | TGD · Sparse CCA misspecified Identity (300,300,500) | prediction loss: 0.0190 -> 0.0113 | No clear effect |
Closest papers by meaning
- Efficient Algorithms for Large-scale Generalized Eigenvector Computation and Canonical Correlation Analysis (ICML 2016 · similarity 0.89)
- On Sparse Canonical Correlation Analysis (NeurIPS 2024 · similarity 0.89)
- Sparse PCA from Sparse Linear Regression (NeurIPS 2018 · similarity 0.88)
- Efficient Sparse PCA via Block-Diagonalization (ICLR 2025 · similarity 0.86)
- Fast and Provable Algorithms for Sparse PCA with Improved Sample Complexity (ICML 2025 · similarity 0.86)
- Unconstrained Stochastic CCA: Unifying Multiview and Self-Supervised Learning (ICLR 2024 · similarity 0.85)
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.
