Call for Papers

Workshop Scope

Scientific imaging and spatiotemporal datasets—from medical imaging, climate and weather, earth systems, materials science, manufacturing, bioengineering, microscopy, cosmology and nuclear fusion—are rapidly increasing in spatial resolution, temporal duration, modality, and dimensionality. These datasets now exceed the capability of traditional AI workflows developed for commodity-scale computing.

Leadership-class HPC systems make it possible to train and deploy AI models that process unprecedented spatial and temporal scales, integrate heterogeneous data sources, and operate at scientific fidelity. Such capabilities are enabling scientific analyses that were previously computationally infeasible, including ultra-high-resolution digital twins, long-horizon forecasting, multimodal scientific foundation models, uncertainty-aware inference, and real-time decision support for experiments and simulations.

Scale-to-See establishes a dedicated forum at the intersection of high-performance computing (HPC), artificial intelligence (AI), scientific imaging, and high-dimensional spatiotemporal data. While AI for science, large-scale simulation, and imaging are each active areas, there is currently no cohesive venue focused on AI methods and systems designed to operate at extreme HPC scale for imaging and spatiotemporal inference.

Scientific imaging and spatiotemporal data—arising in domains such as medical imaging, climate and weather, earth systems, materials science, manufacturing, bioengineering, and cosmology—are increasingly high-dimensional, multimodal, and time-evolving. Advancing these fields requires AI that is not only accurate, but also scalable, physics-aware, reproducible, and deployable on leadership-class HPC systems.

This workshop aims to answer this central question:
How can extreme-scale AI transform scientific discovery for imaging and spatiotemporal data applications? 

More specifically:

• How do we design AI algorithms that scale efficiently on leadership-class HPC systems for imaging and spatiotemporal data applications?

• How do we integrate imaging and spatiotemporal data physical principles, scientific knowledge, and uncertainty into these AI models at scale?

• How can scaling AI unlock scientific capabilities that were previously impossible due to computational limitations?

• What new observations, measurements, or discoveries become feasible when processing unprecedented spatial, temporal, and multimodal imaging and spatiotemporal datasets?

Topics

Topics include but are not limited to:

Core Theme 1: Scientific Observation and Computational Imaging

• Computational imaging and inverse problems at scale

• Scientific sensing, imaging, and observation systems at scale

• Large scale scientific imaging reconstruction and enhancement

• Multimodal sensing and data fusion

• Scientific imaging uncertainty quantification with large ensemble size

• High-dimensional spatiotemporal data analysis

Core Theme 2: Scalable AI for Scientific Observation

• Large foundation models for scientific imaging and spatiotemporal data

• Physics-informed machine learning and scientific AI

• Scientific representation learning and self-supervised learning

• Scientific forecasting, state estimation, and data assimilation

• Scalable generative AI for scientific discovery

Core Theme 3: HPC Systems and Algorithms for Imaging and Spatiotemporal Data

• AI-HPC co-design for imaging or spatiotemporal data

• Distributed training and inference

• Exascale AI systems

• Parallel AI algorithms

• Memory-efficient training

• AI compilers and runtime systems

• AI workflow orchestration

• Hardware-software co-design

• Scientific AI benchmarks

We especially welcome HPC and scalable AI methods that enable seeing observations beyond instrument limits for imaging and spatiotemporal applications. 

Submission Format:

• Paper Length: 4–8 pages (excluding references)

• Format: SC Proceedings Template 

• The review process is double-blind, i.e. the names of the authors, reviewers, and area chairs are not revealed to each other. Papers must thus be properly anonymized before submission. 

• Domain conflicts: To avoid conflict of interest among the authors, reviewers and meta-reviewers, all co-author information and a complete and accurate list of domain conflicts must be properly entered in the SC submission site by the submission deadline. 

Paper Publication:

Accepted papers will be published in SC'26 proceedings as workshop papers.