Research map / Learning theory and algorithms / Kernels and regression theory
Kernel similarity matching with Hebbian networks
Kyle Luther, Sebastian Seung
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 nonlinear features directly with Hebbian updates | kernel similarity matching on Half Moons · Half Moons dataset, low output dimensionality | normalized approximation error | Improved |
| use learned features as Nyström landmarks instead of the learned network responses | kernel similarity matching · Half Moons dataset | normalized approximation error | Improved |
| increase output dimensionality (ablation) | Hebbian kernel similarity matching · Half Moons dataset | normalized approximation error | Hurt |
| increase output dimensionality (ablation) | Hebbian kernel similarity matching · MNIST, α=1 and α=3 | normalized approximation error | Hurt |
| use kernel PCA instead of the proposed learned features | kernel approximation on MNIST · MNIST, α=3 | normalized approximation error | Improved |
| use Nyström landmarks from KMeans or learned features rather than sampled landmarks | kernel approximation on MNIST · MNIST, α=3 | normalized approximation error | Improved |
| use the linear kernel (α=1) rather than nonlinear power-cosine kernels (ablation) | Hebbian kernel similarity matching · MNIST | linear classifier test accuracy | Mixed |
| use nonlinear power-cosine kernels (α=2, 3, or 4) rather than the linear kernel (ablation) | Hebbian kernel similarity matching · MNIST | linear classifier test accuracy | Improved |
| increase kernel exponent from α=1 toward α=2, 3, or 4 (ablation) | Hebbian kernel similarity matching · MNIST | response sparsity | Improved |
| use a higher-dimensional random Fourier representation | Gaussian-kernel approximation · Half Moons dataset | normalized approximation error | Improved |
| use a large ratio of feature and regulatory learning rates to lateral learning rate (ablation) | Hebbian kernel similarity matching · training convergence | convergence | Hurt |
Closest papers by meaning
- Neural Networks as Kernel Learners: The Silent Alignment Effect (ICLR 2022 · similarity 0.86)
- Similarity Learning via Kernel Preserving Embedding (AAAI 2019 · similarity 0.86)
- Spectral Analysis of Representational Similarity with Limited Neurons (NeurIPS 2025 · similarity 0.84)
- On the Target-kernel Alignment: a Unified Analysis with Kernel Complexity (NeurIPS 2024 · similarity 0.84)
- Scalable Neural Network Kernels (ICLR 2024 · similarity 0.84)
- When Can We Approximate Wide Contrastive Models with Neural Tangent Kernels and Principal Component Analysis? (AAAI 2025 · 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.
