Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Scalable Multi-Class Gaussian Process Classification using Expectation Propagation
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 EP instead of VI | Multi-class GP classification · Glass, Svmguide2, Waveform (UCI; M = 20%) | negative test log-likelihood: Glass: 2.30 ± 0.15 -> 0.75 ± 0.06; Svmguide2: 0.94 ± 0.08 -> 0.65 ± 0.06; Waveform: 0.80 ± 0.01 -> 0.37 ± 0.01 | Helped |
| Use SEP, tying the approximate factors, instead of EP (ablation) | Proposed multi-class GP classifier · UCI datasets | training time: between 2 and 3 times faster than GFITC | Helped |
| Use EP rather than GFITC | Multi-class GP classification · UCI datasets | negative test log-likelihood | Mixed |
| Use EP or SEP rather than VI | Multi-class GP classification with stochastic gradients · MNIST | time to convergence | Helped |
| Use SEP rather than VI | Multi-class GP classification · MNIST | test error: 2.08% vs 2.02% | Mixed |
| Use SEP rather than VI | Multi-class GP classification · MNIST | negative test log-likelihood: 0.0725 vs 0.0686 | Mixed |
| Update approximate factors and hyperparameters together instead of rerunning EP to convergence for every hyperparameter update (ablation) | EP hyperparameter optimization · Vehicle | training efficiency | Helped |
| Use the approximate gradient instead of the exact gradient for inner updates (ablation) | EP hyperparameter optimization · Vehicle | marginal-likelihood optimization result | No clear effect |
| Tie all approximate factors using SEP rather than storing them individually (ablation) | EP · UCI datasets | negative test log-likelihood | No clear effect |
| Use VI rather than EP/SEP for inducing-point learning | Sparse multi-class GP classifier · Synthetic two-dimensional three-class task | decision-boundary accuracy with few inducing points | Hurt |
| Increase the number of inducing points | Sparse multi-class GP classifiers · Synthetic two-dimensional three-class task | classification accuracy | Helped |
| Use mini-batches and stochastic gradients | Proposed EP/SEP classifier · Large datasets | scalability to dataset size | Helped |
Closest papers by meaning
- Efficient Transformed Gaussian Processes for Non-Stationary Dependent Multi-class Classification (ICML 2023 · similarity 0.87)
- GP-Tree: A Gaussian Process Classifier for Few-Shot Incremental Learning (ICML 2021 · similarity 0.87)
- KernelMatmul: Scaling Gaussian Processes to Large Time Series (AAAI 2025 · similarity 0.84)
- Learning Mixtures of Gaussian Processes through Random Projection (ICML 2024 · similarity 0.84)
- On the Consistency of Kernel Methods with Dependent Observations (ICML 2024 · similarity 0.84)
- SimPro: A Simple Probabilistic Framework Towards Realistic Long-Tailed Semi-Supervised Learning (ICML 2024 · 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.