Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Healing Products of Gaussian Process Experts
Samuel Cohen, Rendani Mbuvha, Tshilidzi Marwala, Marc Deisenroth
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 uniform expert weights with normalized softmax weights based on predictive variance | gPoE/rBCM · Regression benchmarks, clustering partitioning (average) | NLPD: 0.923 -> 0.354 | Improved |
| Combine experts using a Wasserstein barycenter with softmax-variance weights | BAR_var · Regression benchmarks, clustering partitioning (average) | NLPD: 0.923 -> 0.353 | Improved |
| Use softmax-variance rather than differential-entropy weights | gPoE/rBCM · MNIST PCA features | NLPD: 0.878 -> 0.309 | Improved |
| Use softmax-variance rather than uniform weights | gPoE · MNIST PCA features | NLPD: 0.382 -> 0.309 | Improved |
| Use temperature T=1 rather than T=100 for variance-softmax weights (ablation) | BAR_var · Airfoil | NLPD: 0.411 -> 1.219 | Hurt |
| Use temperature T=1 rather than T=100 for variance-softmax weights (ablation) | gPoE_var · Airfoil | NLPD: 0.411 -> 0.510 | Hurt |
| Increase the temperature from T=1 to T=100 (ablation) | rBCM_var with unnormalized weights · Airfoil | NLPD: 0.412 -> 0.833 | Hurt |
| Increase points per expert from 100 to 500 (ablation) | BAR_var · Kin40k | NLPD: -0.365 -> -0.720 | Improved |
| Increase points per expert from 100 to 500 (ablation) | gPoE_var · Kin40k | NLPD: -0.359 -> -0.765 | Improved |
| Increase points per expert from 100 to 500 (ablation) | rBCM_var · Kin40k | NLPD: 0.124 -> -0.679 | Improved |
| Use normalized rather than unnormalized softmax weights | rBCM · Regression benchmarks across temperatures | NLPD | Improved |
| Aggregate predictions in f-space rather than y-space | Product-of-experts models · Regression benchmarks | NLPD | Improved |
Closest papers by meaning
- Robust and Computation-Aware Gaussian Processes (NeurIPS 2025 · similarity 0.86)
- Robust Gaussian Processes via Relevance Pursuit (NeurIPS 2024 · similarity 0.86)
- Robust and Conjugate Gaussian Process Regression (ICML 2024 · similarity 0.85)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.85)
- Computation-Aware Gaussian Processes: Model Selection And Linear-Time Inference (NeurIPS 2024 · similarity 0.85)
- Bayesian Optimization via Continual Variational Last Layer Training (ICLR 2025 · similarity 0.84)
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.
