Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
SKIing on Simplices: Kernel Interpolation on the Permutohedral Lattice for Scalable Gaussian Processes
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 |
|---|---|---|---|
| Use Simplex-GP instead of SKIP | GP regression · Precipitation | test RMSE: 1.032 ± 0.001 -> 0.939 ± 0.001 | Helped |
| Use Simplex-GP instead of SKIP | GP regression · Keggdirected | test RMSE: 0.487 ± 0.005 -> 0.095 ± 0.002 | Helped |
| Use Simplex-GP instead of SKIP | GP regression · Protein | test RMSE: 0.817 ± 0.012 -> 0.571 ± 0.003 | Helped |
| Use Simplex-GP instead of SGPR | GP regression · Elevators | test RMSE: 0.356 ± 0.006 -> 0.510 ± 0.018 | Hurt |
| Use Simplex-GP instead of SKIP | GP regression · Elevators | test NLL: 0.869 ± 0.074 -> 1.600 ± 0.020 | Hurt |
| Increase blur stencil order from r=1 to r=2 or r=3 (ablation) | Simplex-GP · UCI regression datasets | test RMSE | No clear effect |
| Reduce CG training tolerance from 1.0 to 10^-4 (ablation) | Simplex-GP · Keggdirected training | test RMSE stability | Helped |
| Reduce CG training tolerance from 10^-2 to 10^-4 (ablation) | Simplex-GP · Houseelectric | training runtime per epoch (seconds): 450-600 -> 2000-4000 | Hurt |
| Use RR-CG instead of CG at tolerance 10^-4 (ablation) | Simplex-GP · Houseelectric | training runtime per epoch (seconds): 2000-4000 -> 950-1200 | Helped |
Closest papers by meaning
- Kernel Interpolation with Sparse Grids (NeurIPS 2022 · similarity 0.93)
- The Price of Linear Time: Error Analysis of Structured Kernel Interpolation (ICML 2025 · similarity 0.91)
- gp2Scale: A Class of Compactly Supported Non-Stationary Kernels and Distributed Computing for Exact Gaussian Processes on 10 Million Data Points (ICML 2026 · similarity 0.87)
- Scalable Gaussian Processes with Grid-Structured Eigenfunctions (GP-GRIEF) (ICML 2018 · similarity 0.87)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.85)
- Scalable Derivative Gaussian Processes via Exact Gradient Reduction (NeurIPS 2026 · 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.