Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Meta-Learning Stationary Stochastic Process Prediction with Convolutional Neural Processes
Andrew Foong, Wessel Bruinsma, Jonathan Gordon, Yann Dubois, James Requeima, Richard Turner
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 |
|---|---|---|---|
| train with maximum-likelihood objective L_ML instead of ELBO-style L_NP (ablation) | ConvNP · MNIST image completion | log-likelihood (lower bound for NPs): 0.99 -> 2.11 | Improved |
| train with L_ML instead of L_NP (ablation) | ConvNP · CelebA32 image completion | log-likelihood (lower bound for NPs): -0.27 -> 6.92 | Improved |
| train with L_ML instead of L_NP (ablation) | ConvNP · ZSMM image completion | log-likelihood (lower bound for NPs): 0.14 -> 4.58 | Improved |
| train with L_ML instead of L_NP (ablation) | ANP · 1D regression Matern-5/2 within range | log-likelihood (lower bound for NPs): -0.96 -> -0.73 | Improved |
| train with L_ML instead of L_NP (ablation) | ANP · image completion | log-likelihood (lower bound for NPs): 1.64 -> 1.66 on MNIST; 6.04 -> 5.98 on CelebA32 | Mixed |
| add translation equivariance via convolutional architecture (ConvNP vs ANP) | ANP · ZSMM image completion | log-likelihood (lower bound for NPs): -10.8 -> 4.58 | Improved |
| add translation equivariance (ConvNP vs ANP) beyond training range | ANP · 1D regression sawtooth beyond range | log-likelihood (lower bound for NPs): -0.17 -> 2.29 | Improved |
| add correlated latent-variable predictive (ConvNP vs factorized ConvCNP) | ConvCNP · 1D regression Matern-5/2 within range | log-likelihood (lower bound for NPs): -0.88 -> -0.58 | Improved |
| use ConvNP instead of per-task GP | GP · ERA5-Land Central (train) | joint log-likelihood: 3.33 -> 4.47 | Improved |
| use ConvNP instead of per-task GP | GP · ERA5-Land RMSE | RMSE: 7.94 -> 6.92 on South; 3.10 vs 3.23 on East | Mixed |
| use ConvNP Thompson sampling or UCB instead of GP | GP · ERA5 Bayesian optimization | average regret | Improved |
Closest papers by meaning
- Stochastic Process Learning via Operator Flow Matching (NeurIPS 2025 · similarity 0.83)
- Transformer Neural Processes: Uncertainty-Aware Meta Learning Via Sequence Modeling (ICML 2022 · similarity 0.83)
- On Contrastive Representations of Stochastic Processes (NeurIPS 2021 · similarity 0.82)
- Neural Diffusion Processes (ICML 2023 · similarity 0.82)
- Deep Stochastic Processes via Functional Markov Transition Operators (NeurIPS 2023 · similarity 0.82)
- Flow Matching Neural Processes (NeurIPS 2025 · similarity 0.80)
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.
