AI Paper Atlas

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

NeurIPS 2020 · Probabilistic and scientific modeling › Uncertainty and Gaussian processes · approach: Infinite-width networks as Gaussian processes · 18 citations (OpenAlex)

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 changedWhereMeasuredEffect
train with maximum-likelihood objective L_ML instead of ELBO-style L_NP (ablation)ConvNP · MNIST image completionlog-likelihood (lower bound for NPs): 0.99 -> 2.11Improved
train with L_ML instead of L_NP (ablation)ConvNP · CelebA32 image completionlog-likelihood (lower bound for NPs): -0.27 -> 6.92Improved
train with L_ML instead of L_NP (ablation)ConvNP · ZSMM image completionlog-likelihood (lower bound for NPs): 0.14 -> 4.58Improved
train with L_ML instead of L_NP (ablation)ANP · 1D regression Matern-5/2 within rangelog-likelihood (lower bound for NPs): -0.96 -> -0.73Improved
train with L_ML instead of L_NP (ablation)ANP · image completionlog-likelihood (lower bound for NPs): 1.64 -> 1.66 on MNIST; 6.04 -> 5.98 on CelebA32Mixed
add translation equivariance via convolutional architecture (ConvNP vs ANP)ANP · ZSMM image completionlog-likelihood (lower bound for NPs): -10.8 -> 4.58Improved
add translation equivariance (ConvNP vs ANP) beyond training rangeANP · 1D regression sawtooth beyond rangelog-likelihood (lower bound for NPs): -0.17 -> 2.29Improved
add correlated latent-variable predictive (ConvNP vs factorized ConvCNP)ConvCNP · 1D regression Matern-5/2 within rangelog-likelihood (lower bound for NPs): -0.88 -> -0.58Improved
use ConvNP instead of per-task GPGP · ERA5-Land Central (train)joint log-likelihood: 3.33 -> 4.47Improved
use ConvNP instead of per-task GPGP · ERA5-Land RMSERMSE: 7.94 -> 6.92 on South; 3.10 vs 3.23 on EastMixed
use ConvNP Thompson sampling or UCB instead of GPGP · ERA5 Bayesian optimizationaverage regretImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy