ABSTRACT. As the use of artificial intelligence (AI) and machine learning (ML) is expanding and disseminating for atomistic modeling applications, valid concerns have been recently raised with respect to the amount of task-specific data and computational resources needed to develop AI/ML models that achieve desired accuracy. In response, a new paradigm has been adopted to develop graph foundation models (GFMs) for atomistic modeling applications in two steps. The first step, called pre-training, consists in training a graph neural network (GNN) model on large volumes of generic, task-agnostic atomistic data using supercomputing facilities.

This tutorial aims at covering all the computational challenges that need to be addressed for the pre-training of GFMs, which includes: (1) data gathering, cleaning and pre-processing, (2) efficient scalable data management, (3) design of GNN architecture (4) ensured portability across heterogeneous hardware on different DOE supercomputing facilities (5) stable GFM pre-training on large volumes of multi-source, multi-fidelity data, (6) scalable hyper-parameter optimization (HPO), (7) use of profiling tools to measure the energy consumption for each HPO trial (8) selection of HPO trials to further refine with additional pre-training iterations (9) epistemic uncertainty quantification, (10) open-source release of pre-trained GFMs to the community.

Our tutorial will focus on the technical challenges that we faced, addressed, and solved in steps (1)-(10) across the three main DOE supercomputing facilities, namely NERSC-Perlmutter, ALCF-Aurora, OLCF-Frontier.

ABSTRACT. Cryptographic systems form the bedrock of the modern information age. The ability to establish secure communication on the web is made possible with public key infrastructure. This brief lecture will touch on the practical applications of certificate chains, hashing algorithms, transport layer security (TLS), and hardware backed ssh keys.

ABSTRACT. The Climate-Hydro Analytics Platform (CHAP), developed by Oak Ridge National Laboratory (ORNL), is an innovative, web-based open-access geospatial application designed to provide comprehensive access to high-resolution climate and hydrologic data across the Continental United States (CONUS). As changing environment increasingly impacts water resources, CHAP offers critical support for researchers, policymakers, and water managers by enabling detailed analysis of shifting hydrologic patterns. The platform integrates multiple climate and hydrologic models to project future conditions of key hydroclimate variables such as temperature, precipitation, and runoff at various spatial scales ranging from county and watersheds to river basins (Hydrologic Unit Codes). Users can interactively visualize climate projections and hydrologic responses over monthly, seasonal, and annual timescales, facilitating assessment of potential impacts on water resources. The platform supports 100 TB of data with tailored data downloads in various formats, enabling users to define custom geographic areas, select preferred meteorological forcing datasets, and specify desired variables and aggregation methods. By bridging the gap between complex climate science and practical water resource management, CHAP enhances decision-making capacity. Its user-friendly interface and comprehensive datasets make it an essential resource for evaluating future hydrologic conditions, informing energy resilience, infrastructure planning, and developing sustainable water management strategies.

ABSTRACT. The INTERSECT framework allows for an ecosystem of various micro-services to communicate with each other in order to move data and control devices in a smart lab. For users who want to run their Python scripts within in, the conversion can be a daunting challenge. The Automated Control Testbed for Integration, Verification, and Emulation (ACTIVE) framework provides a simple method for deploying a piece of code into a variety of environments in a rigorously defined way. This tutorial will cover how a code can be implemented as an ACTIVE Strategy and run through ACTIVE, switching between local and INTERSECT environments at will so that scientists can easily prepare their algorithms for deployment and confirm that the version in INTERSECT produces the same results as the local version.

ABSTRACT. This tutorial offers a practical introduction to SAGESim (Scalable Agent-Based GPU-Enabled Simulator), a novel, pure-Python agent-based modeling framework that integrates distributed computing with GPU acceleration for high-performance computing systems including Frontier. As research in complex adaptive systems increasingly demands large-scale simulations, SAGESim meets the critical need for efficient frameworks capable of handling massive agent populations. This tutorial will benefit researchers investigating social networks, human mobility, epidemiology, urban dynamics and transportation, and related fields who require scalable agent-based modeling of complex adaptive systems.

We showcase the framework’s versatility through two detailed case studies. The first implements an SIR (Susceptible-Infected-Recovered) epidemiological model with behavioral heterogeneity, where agents have individualized preventative behavior vectors that influence disease transmission dynamics. The second case study models traffic flow on the OSMNx San Francisco road network, representing intersections with agents that use a popularity-based vehicle redistribution mechanism.

Basic Python programming experience is assumed and basic knowledge of distributed computing, GPU programming model, and OLCF resources may be helpful. By the end, readers will be equipped to design, implement, and deploy large-scale agent-based simulations with SAGESim, enabling them to tackle their own large-scale, previously intractable, complex adaptive system simulation challenges.

ABSTRACT. Research Enablement (RE) is a cross-cutting initiative using structured data engineering, automation, and communication to answer the question, "What does ORNL do?" In this tutorial, RE leads Jason Wohlgemuth, NSSD, and Audrey Carson, CCED, will walk researchers through the process of inputting their project data into JSON format so it can be formatted, improved, and ingested as part of what will become the knowledge base of ORNL research activity data. By inputting their project data, researchers will have a persistent page URL at research.ornl.gov, a PDF fact sheet, and potential to easily update, share, and promote their projects in the future.