Research map / Learning theory and algorithms / Matrix and tensor methods
Robust Spectral Detection of Global Structures in the Data by Learning a Regularization
Pan Zhang
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 |
|---|---|---|---|
| learn diagonal regularization X from localized eigenvectors (X-Laplacian) vs. unregularized adjacency matrix (ablation) | adjacency matrix of SBM · SBM n=10000, c=3, q=2 | overlap | Improved |
| add learned regularization X (X-Laplacian) | SBM with extra cliques · SBM n=10000, c=3, q=2, 10 cliques of size 10 | overlap | Improved |
| fixed-form regularizations (rank-one, Bethe Hessian, normalized Laplacian) instead of learned regularization | SBM with extra cliques · SBM with 10 cliques, epsilon>0.15 | overlap | Hurt |
| add learned regularization X on political blogs adjacency matrix | spectral clustering on political blogs · political blogs (1222 labels) | misclassified labels: 83 -> 50 misclassified | Improved |
| learn regularization for Bethe Hessian | Bethe Hessian · community detection with localization | overlap | Improved |
| use learned regularization on pairwise similarity measurement matrix (vs. Bethe Hessian) | sparse pairwise clustering with local noise · ER topology plus 20 noisy local clusters | overlap | Improved |
| learned regularization vs. trimming+SVD and Bethe Hessian for completion initialization | rank-r matrix completion with cliques · 1000x1000 rank-3, ER + 10 size-20 cliques | fraction of samples with RMSE<1e-7 | Improved |
| learned regularization on SBM with degree-corrected power-law degrees | degree-corrected SBM · n=10000, c=3, q=2, power-law exponent -2.5 | overlap | Improved |
| apply learning on tilde A (normalized adjacency) instead of A | X-Laplacian variant · general | overlap | No clear effect |
Closest papers by meaning
- Robust Spectral Clustering with Rank Statistics (JMLR 2024 · similarity 0.88)
- Structured Low-Rank Matrix Factorization: Global Optimality, Algorithms, and Applications (TPAMI 2019 · similarity 0.88)
- Nonlinear Laplacians: Tunable principal component analysis under directional prior information (NeurIPS 2025 · similarity 0.87)
- Structure-Aware Spectral Sparsification via Uniform Edge Sampling (NeurIPS 2025 · similarity 0.86)
- Spectral Estimation with Free Decompression (NeurIPS 2025 · similarity 0.85)
- Exact Community Recovery under Side Information: Optimality of Spectral Algorithms (ICLR 2025 · 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.
