<ul data-eligibleForWebStory="true"><li>Neural Processes (NPs) are models that predict stochastic processes' posterior predictive distribution.</li><li>Modern NPs handle complex applications such as geology, epidemiology, climate, and robotics.</li><li>The scalability of NPs has become crucial due to data-hungry applications.</li><li>A new architecture, Biased Scan Attention Transformer Neural Process (BSA-TNP), is proposed.</li><li>BSA-TNP introduces Kernel Regression Blocks (KRBlocks) and group-invariant attention biases.</li><li>BSA-TNP uses memory-efficient Biased Scan Attention (BSA) for scalability.</li><li>BSA-TNP matches or surpasses the accuracy of top models while training faster.</li><li>It exhibits translation invariance and can learn at multiple resolutions simultaneously.</li><li>BSA-TNP can model processes evolving in space and time and support high dimensional fixed effects.</li><li>The model can perform inference with over 1M test points and 100K context points in under a minute on a single 24GB GPU.</li></ul>

Scalable Spatiotemporal Inference with Biased Scan Attention Transformer Neural Processes

Discover more