Reproduce “Meta-Learning Stationary Stochastic Process Prediction with Convolutional Neural Processes” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Andrew Y. K. Foong, Wessel P. Bruinsma, Jonathan Gordon, Yann Dubois, James Requeima, Richard E. Turner

ConvNP extends convolutional conditional neural processes with a shared latent field and CNN decoder, producing dependent function predictions while retaining translation equivariance. Its Monte Carlo maximum-likelihood objective directly fits predictive distributions. On zero-shot multi-MNIST completion, training on single digits and testing on pairs on a larger canvas, ConvNP reaches a test log-likelihood lower bound of 4.58 ± 0.04, versus −10.8 ± 1.99 for ANP; this comparison does not isolate equivariance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.