Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
The Functional Neural Process
Christos Louizos, Xiahan Shi, Klamer Schutte, Max Welling
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 local latent variables connected through learned dependency graphs instead of global latent variables | FNP · MNIST and CIFAR10 OOD datasets | AUCR: MNIST: 99.8±0.1; CIFAR10: 93.7±1.0 | Improved |
| Add a direct prediction path from input embedding u | FNP · Toy regression | extrapolation | Improved |
| Use FNP+ rather than FNP | FNP · MNIST and CIFAR10 OOD datasets | AUCR: MNIST: 99.8±0.1 -> 99.9±0.1; CIFAR10: 93.7±1.0 -> 94.9±1.6 | Mixed |
| Use a fixed reference set instead of a randomly sampled reference set | NP · MNIST and CIFAR10 OOD datasets | AUCR | Hurt |
| Randomize the reference set across batches | FNP+ · MNIST and CIFAR10 OOD datasets | AUCR | Improved |
| Reduce FNP reference set size from 300 to 50 (ablation) | FNP · MNIST | test error (%): 0.6 -> 30.6 | Hurt |
| Reduce FNP reference set size from 300 to 50 (ablation) | FNP · MNIST OOD datasets | AUCR: 99.8±0.1 -> 80.0±0.3 | Hurt |
| Vary reference set size across 100, 200, and 500 points (ablation) | FNP and FNP+ · MNIST | test error (%): FNP: 0.9, 0.8, 0.9; FNP+: 0.9, 0.7, 0.7 | No clear effect |
| Reduce FNP+ reference set size from 300 to 50 (ablation) | FNP+ · MNIST | AUCR: 99.9±0.1 -> 99.7±0.2 | No clear effect |
| Reduce FNP+ embedding dimension u to 2 and increase z to 94 (ablation) | FNP+ · MNIST | test error (%): 0.7 -> 4.7 | Hurt |
| Increase z capacity relative to u (ablation) | FNP+ · MNIST OOD datasets | average predictive entropy: 1.8±0.1 -> 2.0±0.1 | Improved |
| Increase u capacity relative to z (ablation) | FNP+ · MNIST | test error (%): 0.7 -> 0.7 | No clear effect |
Closest papers by meaning
- Residual Neural Processes (AAAI 2020 · similarity 0.88)
- Neural Diffusion Processes (ICML 2023 · similarity 0.88)
- Functional Variational Inference based on Stochastic Process Generators (NeurIPS 2021 · similarity 0.88)
- Stochastic Process Learning via Operator Flow Matching (NeurIPS 2025 · similarity 0.86)
- Rényi Neural Processes (ICML 2025 · similarity 0.86)
- Distance-informed Neural Processes (NeurIPS 2025 · 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.
