Advances in foundational AI (fAI), especially those driven by deep neural networks, are urgently needed to address the astrophysics challenges posed by large surveys; SkAI’s Astro-AI interdisciplinary teams will build the requisite technical capabilities by pursuing innovations that span three critical fAI areas.
Generative Models will provide a scalable learning paradigm in which the primary objective is to output new samples from a distribution known only via a collection of training samples. Such models be trained without expensive labeling and will fuel many machine-learning tasks. We will expand the current forefront of these approaches by accounting for multi-modal data, mode collapse, and the treatment of rare events critical to discovering astrophysical phenomena. We will also leverage generative models for astronomy tasks such as missing data imputation, image reconstruction, and simulation acceleration.
Overcoming the inscrutable nature of current deep generative models to enable physically interpretable scientific analyses will require fundamental advances in our second fAI area, Astrophysics-Informed and Interpretable Architectures. New techniques for weaving sophisticated astrophysical guidance (not just straightforward symmetries and constraints) into the structure of models are paramount to ensuring that such systems produce physically consistent predictions.
Our third fAI area, Uncertainty Quantification is critical to validate the reliability of model outputs, guide learning with few labels, and derive reliable astrophysical predictions. Distribution-free predictive inference, Bayesian methods, and data assimilation must be integrated into learning systems to generate informative and actionable uncertainty estimates.
SkAI will realize advances in each fAI area to open new paths to answering key astronomy questions. Combined, these areas will advance trustworthy AI systems that leverage domain knowledge and simulations alongside large-scale observational data. The intentional, cross-disciplinary approach adopted by the SkAI Institute will transform discovery, simulations, and experimental design across astrophysics and accelerate advances in other natural sciences.
The SkAI Institute is one of the National Artificial Intelligence Research Institutes funded by the National Science Foundation (NSF) and Simons Foundation.
Information on National AI Institutes is available at aiinstitutes.org