Sparse-Context Science Using Conditional Neural Processes
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 668-671
DOI: 10.1609/aaaiss.v10i1.43472 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Emily Shinkle und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Scientists aspire to understandable modeling, with a strong bias towards Occam's Razor. Data-scaling philosophies and neural network architectures for AI and foundation model applications have spiraled in complexity, yielding unwieldy workflows with many hyperparameters, co-trained models, optimizer schedules, increasing workflow fragility. In this work we revisit a simple alternative to the transformer, the Conditional Neural Process (CNP). We apply these compact, uncertainty-aware meta-learners to two sparse-context scientific problems: forecasting county-level epidemic trajectories and predicting metal-ligand distribution ratios in separation experiments. For epidemiology, we successfully use CNPs for forecasting diverse synthetic epidemiological data, where context is used to infer an embedding of the underlying epidemic dynamics. For separation chemistry, the CNP model shows passable accuracy for predicting metal-ligand distribution ratios that are important to many industrial and scientific tasks.
Zitieren
APA
Shinkle, E., Alexander, P. C., Lubbers, N., & Pimonova, Y. (2026). Sparse-Context Science Using Conditional Neural Processes. Proceedings of the AAAI Symposium Series, 10(1), 668-671. https://doi.org/10.1609/aaaiss.v10i1.43472
Vancouver
Shinkle E, Alexander PC, Lubbers N, Pimonova Y. Sparse-Context Science Using Conditional Neural Processes. Proceedings of the AAAI Symposium Series. 2026;10(1):668-671. doi:10.1609/aaaiss.v10i1.43472
BibTeX
@article{Shinkle2026,
author = {Shinkle, Emily and Alexander, Prescott C. and Lubbers, Nicholas and Pimonova, Yulia},
title = {Sparse-Context Science Using Conditional Neural Processes},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {668--671},
doi = {10.1609/aaaiss.v10i1.43472}
}
EUVIMED