Sparse observations and coarse-resolution climate models limit effective regional decision-making, underscoring the need for robust downscaling. However, existing AI methods struggle with generalization across variables and geographies and are constrained by the quadratic complexity of Vision Transformer (ViT) self-attention. We introduce ORBIT-2, a scalable foundation model for global, hyper-resolution climate downscaling. ORBIT-2 incorporates two key innovations: (1) Residual Slim ViT (Reslim), a lightweight architecture with residual learning and Bayesian regularization for efficient, robust prediction; and (2) TILES, a tile-wise sequence scaling algorithm that reduces self-attention complexity from quadratic to linear, enabling long-sequence processing and massive parallelism. ORBIT-2 scales to 10 billion parameters across 65,536 GPUs, achieving up to 4.1 ExaFLOPS sustained throughput and 74–98% strong scaling efficiency. It supports downscaling to 0.9 km global resolution and processes sequences up to 4.2 billion tokens. On 7 km resolution benchmarks, ORBIT-2 achieves high accuracy with R2 scores in range of 0.98–0.99 against observation data.

Vision-based scientific foundation models hold significant promise for advancing scientific discovery and innovation. This potential stems from their ability to aggregate images from diverse sources—such as varying physical groundings or data acquisition systems—and to learn spatio-temporal correlations using transformer architectures. However, tokenizing and aggregating images can be compute-intensive, a challenge not fully addressed by current distributed methods. In this work, we introduce the Distributed Cross-Channel Hierarchical Aggregation (D-CHAG) approach designed for datasets with a large number of channels across image modalities. Our method is compatible with any model-parallel strategy and any type of vision transformer architecture, significantly improving computational efficiency. We evaluated D-CHAG on hyperspectral imaging and weather forecasting tasks. When integrated with tensor parallelism and model sharding, our approach achieved up to a 75% reduction in memory usage and more than doubled sustained throughput on up to 1,024 AMD GPUs on the Frontier Supercomputer.

While the pretraining of Foundation Models (FMs) for remote sensing (RS) imagery is on the rise, models remain restricted to a few hundred million parameters. Scaling models to billions of parameters has been shown to yield unprecedented benefits including emergent abilities, but requires data scaling and computing resources typically not available outside industry R&D labs. In this work, we pair high-performance computing resources including Frontier supercomputer, America's first exascale system, and high-resolution optical RS data to pretrain billion-scale FMs. Our study assesses performance of different pretrained variants of vision Transformers across image classification, semantic segmentation and object detection benchmarks, which highlight the importance of data scaling for effective model scaling. Moreover, we discuss construction of a novel TIU pretraining dataset, model initialization, with data and pretrained models intended for public release. By discussing technical challenges and details often lacking in the related literature, this work is intended to offer best practices to the geospatial community toward efficient training and benchmarking of larger FMs.

Earth system predictability is challenged by the complexity of environmental dynamics and the multitude of variables involved. Current AI foundation models, although advanced by leveraging large and heterogeneous data, are often constrained by their size and data integration, limiting their effectiveness in addressing the full range of Earth system prediction challenges. To overcome these limitations, we introduce the Oak Ridge Base Foundation Model for Earth System Predictability (ORBIT), an advanced vision transformer model that scales up to 113 billion parameters using a novel hybrid tensor-data orthogonal parallelism technique. As the largest model of its kind, ORBIT surpasses the current climate AI foundation model size by a thousandfold. Performance scaling tests conducted on the Frontier supercomputer have demonstrated that ORBIT achieves 684 petaFLOPS to 1.6 exaFLOPS sustained throughput, with scaling efficiency maintained at 41% to 85% across 49,152 AMD GPUs. These breakthroughs establish new advances in AIdriven climate modeling and demonstrate promise to significantly improve the Earth system predictability.

Archives of remote sensing (RS) data are increasing swiftly as new sensing modalities with enhanced spatiotemporal resolution become operational. While promising new breakthroughs, the sheer volume of RS archives stretches the limits of human analysts and existing AI tools, as most models are: i) limited to single data modalities; ii) task-specific; iii) heavily reliant on labeled data. The emerging Foundation Models (FMs) have the potential to address these limitations. Trained on vast unlabeled datasets through self-supervised learning, FMs enable generic feature extraction that facilitate specialization to a wide variety of downstream tasks. This paper describes a vision towards an FM for multimodal Earth Observation data (FM4EO), discussing key building blocks and open challenges. We put particular emphasis on multimodal reasoning, a topic underexplored in EO. Our ultimate goal is a practical path toward FM4EO with capacity to unlock breakthroughs in few-shot learning scenarios, multimodal geographic knowledge integration, synthesis, and hypothesis generation.

The peculiar nature of whole slide imaging (WSI), digitizing conventional glass slides to obtain multiple high resolution images which capture microscopic details of a patient’s histopathological features, has garnered increased interest from the computer vision research community over the last two decades. Given the unique computational space and time complexity inherent to gigapixel-size whole slide image data, researchers have proposed novel machine learning algorithms to aid in the performance of diagnostic tasks in clinical pathology. One effective algorithm represents a Whole slide image as a bag of smaller image patches, which can be represented as low-dimension image patch embeddings. Weakly supervised deep-learning methods, such as cluster-constrained-attention multiple instance learning (CLAM), have shown promising results when combined with image patch embeddings. While traditional ensemble classifiers yield improved task performance, such methods come with a steep cost in model complexity. Through knowledge distillation, it is possible to retain some performance improvements from an ensemble, while minimizing costs to model complexity. In this work, we implement a weakly supervised ensemble using clustering-constrained-attention multiple-instance learners (CLAM), which uses attention and instance-level clustering to identify task salient regions and feature extraction in whole slides. By applying logit-based and attention-based knowledge distillation, we show it is possible to retain some performance improvements resulting from the ensemble at zero cost to model complexity.

Ptychography is a popular microscopic imaging modality for many scientific discoveries and sets the record for highest image resolution. Unfortunately, the high image resolution for ptychographic reconstruction requires significant amount of memory and computations, forcing many applications to compromise their image resolution in exchange for a smaller memory footprint and a shorter reconstruction time. In this paper, we propose a novel image gradient decomposition method that significantly reduces the memory footprint for ptychographic reconstruction by tessellating image gradients and diffraction measurements into tiles. In addition, we propose a parallel image gradient decomposition method that enables asynchronous point-to-point communications and parallel pipelining with minimal overhead on a large number of GPUs. Our experiments on a Titanate material dataset (PbTiO3) with 16632 probe locations show that our Gradient Decomposition algorithm reduces memory footprint by 51 times. In addition, it achieves time-to-solution within 2.2 minutes by scaling to 4158 GPUs with a super-linear strong scaling efficiency at 364% compared to runtimes at 6 GPUs. This performance is 2.7 times more memory efficient, 9 times more scalable and 86 times faster than the state-of-the-art algorithm.

Studies on domain adaptation (DA) for remote sensing (RS) imagery analysis lack consistency in selection and description of evaluation scenarios. Without properly characterizing datasets, model assumptions, and evaluation scenarios, it is difficult to objectively compare DA methods and reach conclusions about their suitability across different applications. With this motivation, this work seeks to empirically assess to which extent the interaction between data characteristics and model assumptions influence the effectiveness of DA methods. Using the widely explored task of building footprint segmentation as case study, we perform a large-scale study across over 200 domain adaptation scenarios that include variations across view angles, areas observed, and sensors used for data acquisition. Rather than adopting different model architectures or optimization criteria, we contrast the performances of two DA methods based on adversarial learning that differ only in their assumptions about source and target domains. Informed by metadata and data characteristics unveiled using traditional computer vision techniques as well as pre-trained deep models, we provide a detailed meta-analysis of experiments highlighting the importance of accurately considering data assumptions for DA in RS segmentation tasks. As a “cherry-picking” exercise demonstrates, different claims regarding which model is best could be made by selecting different subsets of evaluation scenarios. While well-calibrated assumptions can be beneficial, mismatching assumptions can lead to negative biases in DA applications. This study intends to motivate the community towards more consistent evaluation protocols, while providing recommendations and insights toward creating novel benchmark datasets, documenting data characteristics, application-specific knowledge, and model assumptions.

The drug discovery process currently employed in the pharmaceutical industry typically requires about 10 years and $2–3 billion to deliver one new drug. This is both too expensive and too slow, especially in emergencies like the COVID-19 pandemic. In silico methodologies need to be improved both to select better lead compounds, so as to improve the efficiency of later stages in the drug discovery protocol, and to identify those lead compounds more quickly. No known methodological approach can deliver this combination of higher quality and speed. Here, we describe an Integrated Modeling PipEline for COVID Cure by Assessing Better LEads (IMPECCABLE) that employs multiple methodological innovations to overcome this fundamental limitation. We also describe the computational framework that we have developed to support these innovations at scale, and characterize the performance of this framework in terms of throughput, peak performance, and scientific results. We show that individual workflow components deliver 100 × to 1000 × improvement over traditional methods, and that the integration of methods, supported by scalable infrastructure, speeds up drug discovery by orders of magnitudes. IMPECCABLE has screened ∼ 1011 ligands and has been used to discover a promising drug candidate. These capabilities have been used by the US DOE National Virtual Biotechnology Laboratory and the EU Centre of Excellence in Computational Biomedicine.

Semantic segmentation of images is an important computer vision task that emerges in a variety of application domains such as medical imaging, robotic vision and autonomous vehicles to name a few. While these domain-specific image analysis tasks involve relatively small image sizes (∼ 102 × 102), there are many applications that need to train machine learning models on image data with extents that are orders of magnitude larger (∼ 104 × 104). Training deep neural network (DNN) models on large extent images is extremely memory-intensive and often exceeds the memory limitations of a single graphical processing unit, a hardware accelerator of choice for computer vision workloads. Here, an efficient, sample parallel approach to train U-Net models on large extent image data sets is presented. Its advantages and limitations are analyzed and near-linear strong-scaling speedup demonstrated on 256 nodes (1536 GPUs) of the Summit supercomputer. Using a single node of the Summit supercomputer, an early evaluation of a recently released model parallel framework called GPipe is demonstrated to deliver ∼ 2X speedup in executing a U-Net model with an order of magnitude larger number of trainable parameters than reported before. Performance bottlenecks for pipelined training of U-Net models are identified and mitigation strategies to improve the speedups are discussed. Together, these results open up the possibility of combining both approaches into a unified scalable pipelined and data parallel algorithm to efficiently train U-Net models with very large receptive fields on data sets of ultra-large extent images.