Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Flow-Induced Diagonal Gaussian Processes
Moule Lin, Andrea Patane, Weipeng Jing, Shuhao Guan, Goetz Botterweck
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 |
|---|---|---|---|
| Enlarge inducing matrix from 32x32 to 64x64 (ablation) | FiD-GP ResNet-18 · CIFAR-100 | accuracy: 69.83 -> 73.26 | Improved |
| Enlarge inducing matrix from 64x64 to 128x128 (ablation) | FiD-GP ResNet-18 · CIFAR-100 | accuracy: 73.26 -> 76.11 | Improved |
| Enlarge inducing matrix from 128x128 to 256x256 (ablation) | FiD-GP ResNet-18 · CIFAR-100 | accuracy: 76.11 -> 77.48 | Improved |
| Enlarge inducing matrix from 32x32 to 256x256 (parameter compression lost) (ablation) | FiD-GP ResNet-18 · CIFAR-100 | parameters: 1.35M -> 12.1M | Hurt |
| Apply FiD-GP to all layers instead of 4 layers (Matheron) (ablation) | FiD-GP ResNet-18 · ImageNet-1k | parameters (M): 8.48M -> 5.62M | Improved |
| Apply FiD-GP to all layers instead of 4 layers (Matheron) (ablation) | FiD-GP ResNet-18 · ImageNet-1k | accuracy: 70.17 -> 70.19 | No clear effect |
| Apply FiD-GP to all layers instead of 4 layers (Matheron) (ablation) | FiD-GP ResNet-18 · CIFAR-100 | accuracy: 76.27 -> 76.11 | No clear effect |
| Use Matheron instead of Reparam sampling (4 layers) (ablation) | FiD-GP ResNet-18 · ImageNet-1k | accuracy: 70.00 -> 70.17 | No clear effect |
| Use Matheron instead of Reparam sampling (4 layers) (ablation) | FiD-GP ResNet-18 · CIFAR-100 | accuracy: 75.99 -> 76.27 | No clear effect |
| Use FiD-GP instead of FFG-U for OoD detection | ResNet-18 · ImageNet-1k vs SVHN | AUROC: 83.7 -> 99.9 | Improved |
| Use FiD-GP instead of BatchEnsemble for OoD detection | ResNet-18 · ImageNet-1k vs SVHN | AUROC: 94.3 -> 99.9 | Improved |
| Use FiD-GP (Matheron, 4 layers) instead of BatchEnsemble for segmentation accuracy | FCN-ResNet50 · CamVID | mIoU: 63.1 -> 62.8 | Hurt |
Closest papers by meaning
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.85)
- Self-Attention through Kernel-Eigen Pair Sparse Variational Gaussian Processes (ICML 2024 · similarity 0.85)
- CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices (NeurIPS 2025 · similarity 0.84)
- Stream-level Flow Matching with Gaussian Processes (ICML 2025 · similarity 0.84)
- Kernelised Normalising Flows (ICLR 2024 · similarity 0.84)
- Domain Invariant Learning for Gaussian Processes and Bayesian Exploration (AAAI 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.
