Skip to Main Content U.S. Department of Energy
Deep Science Agile Initiative

Deep Learning for Scientific Discovery Agile Investment

Over the past decade, multilayer artificial neural networks have experienced a renaissance. These deep neural networks, also known as "deep learning," enable machines to learn and make decisions without being explicitly programmed. Progress has been driven by unprecedented investments by industry, academia, and government. Success has been unprecedented as well: problems previously not considered tractable have been mastered by computers, in some cases exceeding human performance.

Contemporary deep learning has enabled a next generation of artificial intelligence (AI) applications, opening the door to potential breakthroughs in many aspects of our lives. Understanding the capability (and limitations), and improving contemporary artificial intelligence with application to scientific problems, will enable PNNL to advance the frontiers of scientific research and national security.


PNNL will apply deep learning across mission sciences, enabling groundbreaking discoveries and making transformational impacts to accelerate innovation and scientific discovery. The Deep Science Initiative will focus on scientific achievement, capability development, and mission relevance. Because of this investment, we will better leverage our data, computational resources, and scientific talent to increase our research impact.

Technical Approach

We apply deep learning in four mission-relevant areas:

AI Adversaries in Cyber Systems
AI Adversaries in Cyber Systems: develop AI adversaries using generalized adversarial networks and deep reinforcement learning to generate novel attack patterns and more effective defenses, greatly reducing the limitations of existing approaches
Integrating Scientific Knowledge into Deep Neural Networks
Integrating Scientific Knowledge into Deep Neural Networks: increase predictive power and efficiency of computational models by incorporating scientific knowledge into deep neural networks by design
Detecting Breast Cancer by Integrated Omics and MRIs
Detecting Breast Cancer by Integrated Omics and MRIs: predict a patient's outcome given abbreviated breast MRI and patient records, while increasing specificity and maintaining sensitivity
Deep Learning in High Performance Computing
Deep Learning in High Performance Computing: detect if an application is sensitive to soft errors and approximation strategies, and understand the performance and accuracy trade-offs

Further, by investing in numerous additional mission-focused seedling efforts, we will increase the pace and impact of discovery through applying deep learning across the laboratory's missions. These combined efforts will help make PNNL one of the premier laboratories for leveraging deep learning to advance the frontiers of science and security.

Deep Science Agile Investment Flyer

Deep Learning