Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Spherical Inducing Features for Orthogonally-Decoupled 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 |
|---|---|---|---|
| add K=128 orthogonal inducing variables to ReLU/SOFTPLUS activated SVGP (ablation) | activated SVGP with spherical NN features · UCI regression (yacht, concrete, energy, kin8nm, power) | test RMSE and NLPD | Helped |
| add K=128 orthogonal inducing variables to ReLU activated SVGP (vs. M=128 alone) (ablation) | ReLU activated SVGP, Arccos kernel · UCI concrete | test RMSE: 6.82 -> 5.93 | Helped |
| add K=128 orthogonal inducing variables to ReLU activated SVGP (ablation) | ReLU activated SVGP, Arccos kernel · UCI energy | test RMSE: 1.36 -> 0.47 | Helped |
| add K=128 orthogonal inducing variables to ReLU activated SVGP (ablation) | ReLU activated SVGP, Arccos kernel · UCI kin8nm | test RMSE: 0.12 -> 0.08 | Helped |
| add K=128 orthogonal inducing variables to Arccos SOFTPLUS activated SVGP (ablation) | SOFTPLUS activated SVGP · UCI concrete | test RMSE: 6.69 -> 6.06 (vs M=128 K=0) | Helped |
| add orthogonal inducing variables to activated SVGP on the Arccos kernel with ReLU/SOFTPLUS features | activated SVGP on UCI power dataset · UCI power | test RMSE | Mixed |
| use Matern/SE kernels with spherical features plus orthogonal variables | activated SVGP with Matern-5/2 kernel · UCI regression | test RMSE and NLPD | No clear effect |
| add 1000 base inducing variables instead of 500 (standard SVGP, airline delays) | mini-batch SVGP with Arccos kernel · 2008 airline delays 1M subset | test RMSE and NLPD | Helped |
| double number of SOFTPLUS activated features from 500 to 1,000 (airline delays) | activated SVGP with SOFTPLUS features · 2008 airline delays 1M subset | test RMSE and NLPD | No clear effect |
| add 500 orthogonal inducing variables to SOFTPLUS activated SVGP with 500 features | activated SVGP with SOFTPLUS features · 2008 airline delays 1M subset | test RMSE and NLPD | Helped |
| increase truncation level L from 8 to 16 in activated SVGP (ablation) | activated SVGP Arccos-SOFTPLUS on 1D Snelson · synthetic 1D Snelson | ELBO / fit quality | Mixed |
Closest papers by meaning
- Sparse Gaussian Processes with Spherical Harmonic Features (ICML 2020 · similarity 0.92)
- Orthogonally Decoupled Variational Gaussian Processes (NeurIPS 2018 · similarity 0.89)
- Inter-domain Deep Gaussian Processes (ICML 2020 · similarity 0.89)
- Input Dependent Sparse Gaussian Processes (ICML 2022 · similarity 0.88)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.87)
- Gaussian Process Neural Additive Models (AAAI 2024 · similarity 0.87)
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.