Reproduce “Automatic Integration for Spatiotemporal Neural Point 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
Zihao Zhou, Rose Yu
Researchers at UC San Diego developed AutoSTPP, a deep learning framework that extends automatic integration to 3D spatiotemporal neural point processes, enabling exact likelihood computation. This method utilizes a novel "ProdNet" architecture to enforce non-negativity of the intensity function and demonstrated superior accuracy in recovering complex spatiotemporal dynamics and higher predictive performance on real-world datasets compared to state-of-the-art baselines.
Key insights
- Automatic integration can be successfully extended to 3D spatiotemporal point processes by carefully addressing the non-negativity constraint.
- The ProdNet architecture allows for flexible modeling of non-separable, complex multivariate functions while ensuring non-negative intensity.
- Theoretical proofs confirm AutoSTPP's universal approximation capability for derivative networks and its consistency as an estimator under mild assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.