Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Gaussian RBF Centered Kernel Alignment (CKA) in the Large-Bandwidth Limit
Sergio A. Álvarez
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 |
|---|---|---|---|
| remove mean-centering of Gram matrices | Gaussian RBF CKA · 2x2 counterexample | convergence to linear CKA: Gaussian CKA 1 vs linear 3/sqrt(14) | Hurt |
| use bandwidth sigma larger than rho | Gaussian RBF CKA · all data sets | mean relative difference to linear CKA: less than 0.01 | Improved |
| use small bandwidth sigma <= 0.25 | Gaussian RBF CKA · classification data sets | relative difference to linear CKA: greater than 0.2; greater than 0.7 for dna at w>=128 | Improved |
| increase network width | MLP representations · regression data sets | representation eccentricity rho: cpu 10.63 -> 8.82 | Improved |
| increase network width | MLP representations on balloon · balloon | representation eccentricity rho: 14.62 -> 25.18 | Hurt |
| increase network width | MLP representations · classification data sets | standard error of relative CKA difference | Improved |
| increase rho | Gaussian CKA · data sets | convergence onset log2 sigma*_0: balloon rho 16 -> onset 5; wdbc rho 11 -> 4; cpu rho 9 -> 4 | Hurt |
| raise raw data dimensionality | MLP representations · classification and regression | correlation with rho: -0.38 to -0.33 classification; -0.8 to -0.85 regression | Improved |
Closest papers by meaning
- Reliability of CKA as a Similarity Measure in Deep Learning (ICLR 2023 · similarity 0.83)
- On the Target-kernel Alignment: a Unified Analysis with Kernel Complexity (NeurIPS 2024 · similarity 0.83)
- Enhancing Diversity in Bayesian Deep Learning via Hyperspherical Energy Minimization of CKA (NeurIPS 2024 · similarity 0.83)
- The Conditional Lucas & Kanade Algorithm (ECCV 2016 · similarity 0.82)
- Random features for Grassmannian kernel approximation with bounded rank-one projections (NeurIPS 2026 · similarity 0.81)
- Learning Input Encodings for Kernel-Optimal Implicit Neural Representations (ICML 2025 · similarity 0.81)
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.
