AI Initiative

ORNL Featured Speakers: Ben Mintz and Prasanna Balaprakash
When: Tuesday @ 11:45
Title: Building Assured and Efficient AI-Driven Scientific Laboratories of the Future

Brief Description: The AI and INTERSECT initiatives at Oak Ridge National Laboratory are driving the next generation of AI-enabled, interconnected, and intelligent laboratories. Together, these efforts are advancing scientific AI models and end-to-end autonomous science workflows that integrate AI agents with experimental instruments, edge computing platforms, and high-performance computing (HPC) systems—while also ensuring the trustworthiness and assurance of AI systems. The AI Initiative focuses on the development of assured and efficient scientific AI methods that support hypothesis generation, surrogate modeling to accelerate scientific simulations, adaptive experimental design, and real-time decision-making. In parallel, INTERSECT provides the unified infrastructure and cross-domain connectivity needed to deploy these capabilities across a broad range of facilities and scientific disciplines. This presentation will highlight several scientific AI modalities and autonomous workflows, with a focus on how AI agents are being integrated into real-world applications across materials science, biology, chemistry, and manufacturing.

Amir Sadovnik
When: Tuesday @ 10:00
Demo Title: Red Teaming AI: Demonstrating Tools for AI Vulnerability Assessment 

Brief description: The integration of AI into more national security systems and processes can introduce vulnerabilities. In this demo we will explore the different tools developed at the Center for AI Security Research (CAISER) at Oak Ridge National Laboratory (ORNL) to efficiently detect vulnerabilities in computer vision models. We present an overview of adversarial AI attacks and demonstrate how these tools can be used to assess models for different national security tasks. 

Veronica Melesse Vergarra
When: Tuesday @ 3:15
Demo Title: DOE-NIH-NSF Collaboration: Deploying Biomedical RAG pipelines on Frontier as part of the NAIRR Secure Pilot      

Brief description: This demonstration will showcase new capabilities being piloted by the Oak Ridge Leadership Computing Facility (OLCF) to support the deployment of biomedical RAG pipelines on ORNL's Frontier Citadel environment. This demonstration features work resulting from a collaboration between ORNL, NIH's National Center for Advancing Translational Sciences (NCATS), and NSF as part of the NAIRR Secure Pilot program. The demonstration showcases a query-response user workflow through JupyterHub on OLCF's Frontier supercomputer and a prototype "Ask AIthena" chat agent running on NCATS's DALI system. 

Prasanna Date and Kathleen Hamilton
When: Wednesday @ 11:15
Demo Title: AI and Quantum: Building A Mutually Beneficial Relationship

Brief description: In this demo, we explore two use cases that show how today's AI and quantum computing can benefit each other. First, we will demonstrate how adiabatic quantum computers can be used for training widely used machine learning models such as linear regression, support vector machine (SVM), and k-means clustering. We will show that adiabatic quantum computers can accelerate the training of these models on larger sized datasets containing millions of data points or features. Next, we will demonstrate how AI techniques can benefit quantum applications executed on noisy quantum hardware. Today's gate-based quantum hardware is becoming more advanced, but still prone to noise. Therefore, we need methods that can recover useful information from noisy measurements with minimal data overhead. We will demonstrate how AI-based denoising techniques and self-supervised learning can mitigate the effects of noise, getting us closer to utility-scale quantum computing.

Max Lupo Pasini
When: Wednesday @ 2:30
Demo Title: Scalable and Trustworthy Graph Foundation Models for Atomistic Materials Science

Brief Description: We will showcase Oak Ridge National Laboratory’s distinctive strengths at the confluence of high-performance computing, artificial intelligence, and materials science. Leveraging the combined computational power of the Department of Energy’s flagship supercomputers—NERSC Perlmutter, ALCF Aurora, and OLCF Frontier—we will demonstrate how HydraGNN, our in‑house scalable graph‑neural‑network framework, can be applied to develop, train, and validate graph‑based foundation models on hundreds of millions of atomistic structures drawn from diverse, multi‑fidelity first‑principles datasets.

VIDEOS at the AI Expo

• AI and Quantum
• Autoflows for AI
• Autonomous Chemistry Lab
• Frontier AI Capabilities
• INTERSECT
• AI and Additive Manufacturing
• Hurricane Ida Drone Survey

1000 Scientist AI Jam

February 28, 2025

The Department of Energy national laboratories are working together to use artificial intelligence (Al) models for scientific research.

A better understanding and exploring the potential impact of AI reasoning models on national security and science and, in particular, how reasoning models may accelerate science more broadly is of national interest. Toward that end, the DOE national laboratories, in concert with OpenAI, are convening an “AI Jam Session” for scientists to explore the potential of multiple advanced AI models to tackle challenging scientific problems. This will be the first in a series of AI Jam Sessions involving multiple national l laboratories to gather data on model performance across a diverse range of realistic, representative tasks in scientific research and development. The outputs will serve as a first-pass proxy to extrapolate the potential impact of these tools for the scientific community.

December 19, 2024

The inaugural AI initiative’s AI Clinic on December 19, 2024 showcased how two Frontier AI models—OpenAI’s o1 preview and Anthropic’s Claude 3.5 Sonnet—can accelerate scientific research. Over the course of the day, interdisciplinary teams of over 45 members tackled complex challenges, ranging from advanced materials research to climate modeling, into manageable AI-driven subproblems. Teams then presented their findings, demonstrating how tailored AI applications can enhance problem-solving, innovative approaches, and guide future research directions.