Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Scalable Variational Gaussian Processes via Harmonic Kernel Decomposition
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 |
|---|---|---|---|
| Decompose the kernel into orthogonal components and use separate inducing-point groups | Variational Gaussian processes · Earth elevation test | NLL: -0.203 -> -0.314 | Helped |
| Decompose the kernel and use 13 groups of 1k inducing points | Variational Gaussian processes · Earth elevation test | RMSE: 0.145 -> 0.177 | Hurt |
| Use four groups rather than two groups of inducing points (ablation) | HVGP · CIFAR-10, 4-layer convolutional GP | NLL: 0.60$\pm$0.00 -> 0.58$\pm$0.00 | Helped |
| Increase the number of inducing-point groups from one to four (ablation) | HVGP · CIFAR-10, 3-layer convolutional GP | NLL: 0.73$\pm$0.01 -> 0.58$\pm$0.00 | Helped |
| Use a transformation mismatched to the data distribution instead of translations (ablation) | HVGP · Translate-MNIST | test accuracy | Hurt |
| Share inducing points across decomposed components (ablation) | HVGP · Translate-MNIST | test accuracy | No clear effect |
| Use negations rather than image flips as the transformation (ablation) | HVGP · Flip-MNIST | test accuracy | Hurt |
| Increase the number of inducing-point groups from M to 8xM (ablation) | HVGP · Regression benchmarks | NLL | Helped |
Closest papers by meaning
- Variational Inference for Gaussian Process Models with Linear Complexity (NeurIPS 2017 · similarity 0.91)
- Orthogonally Decoupled Variational Gaussian Processes (NeurIPS 2018 · similarity 0.91)
- Scalable Gaussian Processes with Grid-Structured Eigenfunctions (GP-GRIEF) (ICML 2018 · similarity 0.89)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.88)
- Bayesian Basis Function Approximation for Scalable Gaussian Process Priors in Deep Generative Models (ICML 2025 · similarity 0.88)
- Neighbour-Driven Gaussian Process Variational Autoencoders for Scalable Structured Latent Modelling (ICML 2025 · similarity 0.88)
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.