Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Mixtures of Gaussian Process Experts with SMC^2
Teemu Härkönen, Sara Wade, Kody Law, Lassi Roininen
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 |
|---|---|---|---|
| replace importance sampling with SMC^2 inference (alpha=0.1) | mixture of GP experts · Synthetic 2: discontinuous, heteroskedastic noise | L1 distance: 993.265 -> 509.876 | Improved |
| replace importance sampling with SMC^2 inference (alpha=0.1) | mixture of GP experts · Synthetic 2: discontinuous, heteroskedastic noise | predictive log-likelihood: -258.943 -> -224.955 | Improved |
| replace importance sampling with SMC^2 inference (alpha=0.1) | mixture of GP experts · Synthetic 1: non-stationary, homoskedastic noise | L1 distance: 588.068 -> 375.731 | Improved |
| replace importance sampling with SMC^2 inference (alpha=0.1) | mixture of GP experts · Synthetic 1: non-stationary, homoskedastic noise | predictive log-likelihood: -293.514 -> -296.381 | Hurt |
| use SMC^2-MoE (alpha=0.1) instead of single GP | GP regression · Synthetic 2: discontinuous, heteroskedastic noise | L1 distance: 1078.485 -> 509.876 | Improved |
| use SMC^2-MoE (alpha=0.1) instead of single GP | GP regression · Synthetic 1: non-stationary, homoskedastic noise | L1 distance: 373.676 -> 375.731 | No clear effect |
| use SMC^2-MoE (alpha=0.1) instead of TreedGP | mixture of GP experts · Synthetic 1: non-stationary, homoskedastic noise | L1 distance: 363.797 -> 375.731 | Hurt |
| use SMC^2-MoE (alpha=0.1) instead of TreedGP | mixture of GP experts · Synthetic 2: discontinuous, heteroskedastic noise | L1 distance: 655.436 -> 509.876 | Improved |
| increase Dirichlet concentration alpha from 0.1 to K/2 (ablation) | SMC^2-MoE · Synthetic 2: discontinuous, heteroskedastic noise | L1 distance: 509.876 -> 558.190 | Hurt |
| increase Dirichlet concentration alpha from 0.1 to K/2 (ablation) | SMC^2-MoE · Synthetic 1: non-stationary, homoskedastic noise | L1 distance: 375.731 -> 403.566 | Hurt |
| increase Dirichlet concentration alpha from 0.1 to K/2 (ablation) | SMC^2-MoE · computational cost | cost | Improved |
| use SMC^2-MoE (alpha=0.1) instead of RBCM | GP regression · Synthetic 2: discontinuous, heteroskedastic noise | L1 distance: 1217.425 -> 509.876 | Improved |
Closest papers by meaning
- Embarrassingly Parallel Inference for Gaussian Processes (JMLR 2019 · similarity 0.93)
- Scalable Training of Inference Networks for Gaussian-Process Models (ICML 2019 · similarity 0.89)
- Convergence Rates for Gaussian Mixtures of Experts (JMLR 2022 · similarity 0.89)
- Learning Mixtures of Gaussian Processes through Random Projection (ICML 2024 · similarity 0.88)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.86)
- Solving High Frequency and Multi-Scale PDEs with Gaussian Processes (ICLR 2024 · similarity 0.86)
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.
