Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Embarrassingly Parallel Inference for Gaussian Processes
Michael Minyi Zhang, Sinead A. Williamson
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 |
|---|---|---|---|
| use importance weights instead of uniform weights (ablation) | IS-MOE · Synthetic 100-d GMM inputs (J=10, K=10, B=1000) | test log likelihood: -726.67 -> -354.96 | Improved |
| use importance weights instead of uniform weights (ablation) | IS-MOE · Synthetic 100-d GMM inputs (J=10, K=10, B=1000) | test MSE: 0.19 -> 0.096 | Improved |
| add stochastic approximation (minibatching) (ablation) | IS-MOE · Synthetic 100-d GMM inputs (J=10, K=10, B=1000) | test log likelihood: -423.93 -> -354.96 | Improved |
| add stochastic approximation (minibatching) (ablation) | IS-MOE · Synthetic 100-d GMM inputs (J=10, K=10, B=1000) | test MSE: 0.097 -> 0.096 | No clear effect |
| partition with GMM on covariates instead of random clusters (ablation) | IS-MOE · GMM-generated inputs | test log likelihood: -789.37 -> -429.19 | Improved |
| partition with GMM on covariates instead of random clusters (ablation) | IS-MOE · GMM-generated inputs | test MSE: 0.53 -> 0.14 | Improved |
| partition with GMM instead of random clusters (ablation) | IS-MOE · Uniform-generated inputs | test MSE: 0.62 -> 0.62 | No clear effect |
| increase importance samples J from 1 to about 50 | IS-MOE · Synthetic 10k points | test performance | Improved |
| increase number of blocks K or decrease B and J | IS-MOE · AIRS CO2 | test log likelihood | Hurt |
| use IS-MOE instead of full GP | classification · Parkinsons | AUC: 0.86 -> 0.93 | Improved |
| use IS-MOE instead of full GP | classification · Pima | AUC: 0.83 -> 0.81 | Hurt |
| let BTGP chain converge (J=6000) instead of IS-MOE | non-stationary synthetic · Synthetic non-stationary (J=6000) | wall time (s): 61.16 -> 1116.75 | Hurt |
Closest papers by meaning
- Mixtures of Gaussian Process Experts with SMC^2 (JMLR 2025 · similarity 0.93)
- Generic Inference in Latent Gaussian Process Models (JMLR 2019 · similarity 0.89)
- Efficiently sampling functions from Gaussian process posteriors (ICML 2020 · similarity 0.88)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.87)
- Learning Mixtures of Gaussian Processes through Random Projection (ICML 2024 · similarity 0.86)
- Solving High Frequency and Multi-Scale PDEs with Gaussian Processes (ICLR 2024 · 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.
