The following announcement was shared by Lewis-Burke Associates LLC.
On October 22 and 23, the Department of Energy (DOE) Office of Advanced Scientific Computing Research (ASCR) held the fourth and last of a series of townhall meetings to help identify future research directions at the intersection of artificial intelligence (AI) and DOE’s scientific priorities. In total, nearly 1,500 scientists from national laboratories, research universities, and industry participated in all four events. A final report that summarizes the priority research directions and applications of AI to address DOE missions should be completed by the end of the year and will serve as a framework for future investments in DOE’s AI initiative starting in fiscal year (FY) 2021. The summary below provides advance intelligence on priority areas and highlights upcoming FY 2020 funding opportunities in AI.
AI Townhalls and Future Applications
ASCR led these townhalls, and the planning activities that will follow them, as the decade-long Exascale Computing Initiative comes to a close and because the Administration and DOE have identified AI as a major research and development priority. In particular, ASCR views this inflection point as an opportunity to develop a vision for driving advances in AI across the Office of Science that leverages the capabilities of its new exascale systems in concert with a more robust and future-focused research agenda in applied mathematics, computational science, and computer science. The graphic below shows how AI can address DOE’s research needs.
Source: DOE Office of Science.
Ultimately, the findings of these townhalls will be used to map out a strategic vision for leveraging the integration of modeling and simulation, data science, and machine learning toward achieving the following over a 10-year horizon:
- Learned models begin to replace data;
- Experimental discovery processes are dramatically refactored;
- Many questions are pursued semi-autonomously at-scale;
- Simulation and AI approaches merge;
- Theory becomes data for next-generation AI;
- AI becomes a common part of scientific laboratory activities.
Office of Science Director Chris Fall opened the DC-based townhall by outlining in broad terms DOE’s interest in AI and the agency’s rationale for placing itself at the center of a government-wide approach to driving the technology forward. Dr. Fall made the case that as the steward of the National Laboratory complex and the owner of the nation’s most powerful supercomputers, DOE is uniquely positioned to work with other federal agencies as well as external partners to develop AI-enabled applications for significant challenge areas both within and outside of its mission space. He also spoke to the establishment of DOE’s new Artificial Intelligence and Technology Office (AITO), which will support cross-cutting research activities in AI, foster external partnerships, integrate AI capabilities into the agency’s overall operations, and develop ethical and legal policies for the agency’s adoption of AI.
Interagency collaboration and coordination were a major focus of the discussion during the townhall, reflecting DOE’s view that they are essential to its overall vision for AI. As an example, DOE and the National Cancer Institute (NCI) have established the Joint Design of Advanced Computing Solutions for Cancer (JDACS4C) activity, which aims to advance AI-enabled precision oncology by marrying DOE’s computing capabilities and algorithm development with NCI’s biomedical research expertise. JDACS4C is currently limited to three pilot projects with national laboratories, but ASCR is exploring the potential for collaboration with non-federal partners.
All four townhalls featured breakout sessions on AI for specific scientific domains as well as crosscutting areas, though the DC townhall included a session summarizing the findings of the domain-specific breakouts held during the previous three events. AI applications of most interest to DOE include materials, chemistry, nanoscience, Earth systems, biology and life sciences, fundamental physics, engineering manufacturing, smart energy infrastructure, computer science, and fusion. Some of the crosscutting issues being addressed are data life cycle and infrastructure, hardware architectures, AI for experimental facilities, and AI at the edge. In addition, the townhall organizers closed the event by detailing how AI would impact six specific application areas relevant to DOE’s mission:
- Biology: DOE is interested in leveraging AI to accelerate advances in synthetic biology, including through the use of machine learning algorithms to absorb, understand, and manipulate biological processes.
- User Facilities: AI can be used to optimize operations at large scientific user facilities, such as beam lines, and for on-site processing of the data being generated through “AI-at-the-edge” capabilities.
- Materials: DOE is interested in developing machine learning algorithms capable of continually learning from autonomously run experiments and then updating their predictions based on a constant stream of new data. These capabilities may be particularly relevant to the synthesis of quantum materials.
- Cosmology: DOE seeks to use AI to simulate the development of the universe from its formation to the present, and then predict future expansions at all scales. Data generated by current and future dark energy and dark matter experiments will be critical to advances in this application area.
- Manufacturing: AI could be used to optimize supply chains and the synthesis and use of specialty materials, enabling more efficient manufacturing processes.
- Cities: AI can help optimize mobility, safety, energy consumption, and security in cities through the integration of edge computing and sensors.
Additional information about the townhall, including copies of the presentations delivered, can be found here.
FY 2020 AI Funding Opportunities
DOE started making targeted investments in AI in FY 2019, which included funding calls in Data Science for Discovery in Chemical and Materials Sciences, a co-design center for AI hardware and software for cybersecurity and electric grid resilience, and the Advanced Research Project Agency-Energy’s support of AI tools to transform operations and maintenance of advanced nuclear reactors.
DOE plans to build on the abovementioned investments by allocating approximately $123 million to AI and scientific machine learning in FY 2020, including $71 million in the Office of Science, $42 million in the Advanced Scientific Computing program in the National Nuclear Security Administration, and $10 million in the applied energy offices. While a final list of funding opportunities is subject to Congress completing FY 2020 appropriations, DOE plans to release the following funding solicitations focused on AI:
- $10 million for applied mathematics in fundamental scientific machine learning principles identified in the Basic Research Needs report “Scientific Machine Learning: Core Technologies for Artificial Intelligence”;
- $5 million-$10 million for one or two more AI Co-design centers to develop core hardware and software technologies for DOE mission applications;
- $10 million for computational tool development for integrative systems biology data analysis (data integration, analysis and sharing for genomics and synthetic biology);
- $5 million-$10 million for Earth system modeling (computationally advanced climate and Earth system model to investigate the challenges posed by the interactions of climate change with energy and related sectors);
- $7 million for machine learning for fusion energy sciences and plasma physics;
- $13 million for computational materials science centers with an emphasis on data analytics and machine learning for data driven science.