Research map / Learning theory and algorithms / Matrix and tensor methods
A Nonconvex Framework for Structured Dynamic Covariance Recovery
Katherine Tsai, Mladen Kolar, Sanmi Koyejo
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 |
|---|---|---|---|
| add projected-gradient iterative refinement after spectral initialization (ablation) | Dynamic covariance estimation · High-dimensional low-rank simulation, 10 training subjects | average log-Euclidean distance: 0.43\pm0.01 -> 0.42\pm0.03 | Improved |
| add projected-gradient iterative refinement after spectral initialization (ablation) | Dynamic covariance estimation · High-dimensional low-rank simulation, 50 training subjects | average log-Euclidean distance: 0.39\pm0.01 -> 0.32\pm0.02 | Improved |
| add a QR decomposition step (ablation) | Proposed model · High-dimensional low-rank simulation, 10 training subjects | average log-Euclidean distance: 0.42\pm0.03 -> 0.40\pm0.04 | No clear effect |
| add a QR decomposition step (ablation) | Proposed model · High-dimensional low-rank simulation, 50 training subjects | average log-Euclidean distance: 0.32\pm0.02 -> 0.32\pm0.02 | No clear effect |
| use the proposed structured covariance model instead of sliding window shrunk covariance | HCP motor task classification · HCP motor task, 10 training subjects | test classification accuracy: 70.9\pm2.7 -> 61.5\pm7.4 | Hurt |
| use the proposed structured covariance model instead of sparse dictionary learning | HCP motor task classification · HCP motor task, 10 training subjects | test classification accuracy: 52.6\pm8.3 -> 61.5\pm7.4 | Improved |
| use the proposed structured covariance model instead of sparse dictionary learning | HCP motor task classification · HCP motor task, 50 training subjects | test classification accuracy: 90.5\pm2.5 -> 90.5\pm2.5 | No clear effect |
| increase the model rank | Proposed model · High-dimensional low-rank simulation | average log-Euclidean distance | Hurt |
| under-select sparsity level or rank | Proposed model · Dynamic covariance estimation | evaluation score | Hurt |
| vary the kernel length scale and smoothness coefficient | Proposed model · Dynamic covariance estimation | estimation performance | No clear effect |
Closest papers by meaning
- Multivariate Sparse Coding of Nonstationary Covariances with Gaussian Processes (NeurIPS 2019 · similarity 0.82)
- Dynamic matrix recovery from incomplete observations under an exact low-rank constraint (NeurIPS 2016 · similarity 0.80)
- Fast recovery from a union of subspaces (NeurIPS 2016 · similarity 0.80)
- Global Convergence of Sub-gradient Method for Robust Matrix Recovery: Small Initialization, Noisy Measurements, and Over-parameterization (ICML 2023 · similarity 0.80)
- Exploring Invariance in Images through One-way Wave Equations (ICML 2025 · similarity 0.79)
- Improving Diffusion Models for Inverse Problems Using Optimal Posterior Covariance (ICML 2024 · similarity 0.79)
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.
