Research map / Optimization / Optimizers
An Implicit Form of Krasulina's k-PCA Update without the Orthonormality Constraint
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the partially implicit Krasulina update instead of explicit Krasulina | online k-PCA · MNIST and CIFAR-10, k=5, 10, 20 | compression loss: 35.17±0.00 vs 35.17±0.01; 26.96 ± 0.01 vs 26.97± 0.03; 18.79±0.02 vs 18.77± 0.02; 87.04±0.52 vs 87.01± 0.00; 65.90 ± 0.0 | No clear effect |
| Scale initial learning rate to 0.1× or 10× the selected value (ablation) | Implicit Krasulina · MNIST, k=5, 10, 20 | compression loss: 35.17±0.01, 35.17±0.01; 26.96±0.01, 26.98 ± 0.05; 18.78± 0.02, 18.77± 0.01 | No clear effect |
| Scale initial learning rate to 0.1× or 10× the selected value (ablation) | Implicit Krasulina · CIFAR-10, k=5, 10, 20 | compression loss: 87.00± 0.00, 87.02 ± 0.04; 65.75 ± 0.05, 65.74 ± 0.01; 48.75±0.03, 48.76±0.03 | No clear effect |
| Scale initial learning rate away from its selected value (ablation) | Oja and explicit Krasulina · MNIST and CIFAR-10 | compression loss: Oja MNIST k=10: 26.98 ± 0.02 -> 29.78± 0.38; Krasulina CIFAR10 k=5: 87.04±0.52 -> 116.86±2.58 | Hurt |
| Combine 10 implicit Krasulina machine models by averaging | Implicit Krasulina distributed k-PCA · MNIST and CIFAR-10, k=5 and 20 | compression loss: 35.17±0.01 -> 35.17±0.00; 18.77± 0.02 -> 18.77±0.01; 87.01± 0.00 -> 87.02±0.00; 48.74±0.04 -> 48.76±0.04 | No clear effect |
| Use implicit Krasulina rather than the comparison algorithms | online k-PCA · MNIST and CIFAR-10 | convergence/runtime | Helped |
Closest papers by meaning
- Exponentially convergent stochastic k-PCA without variance reduction (NeurIPS 2019 · similarity 0.89)
- Fast Approximate Natural Gradient Descent in a Kronecker Factored Eigenbasis (NeurIPS 2018 · similarity 0.86)
- Diffusion Approximations for Online Principal Component Estimation and Global Convergence (NeurIPS 2017 · similarity 0.86)
- Stochastic PCA with $\ell_2$ and $\ell_1$ Regularization (ICML 2018 · similarity 0.85)
- KOALA++: Efficient Kalman-Based Optimization with Gradient-Covariance Products (NeurIPS 2025 · similarity 0.85)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.84)
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.