The following funding and research opportunities were released by the U.S. Department of Energy and written by Lewis-Burke Associates.
The analysis below provides advance intelligence on upcoming funding opportunities and future research directions for the Department of Energy Office of Science in applied math, computer science, and high performance computing. The analysis is based on information from the March 26-27 Advanced Scientific Computing Research Advisory Committee (ASCAC) meeting and discussions with DOE program managers. This Advisory Committee provides advice to the Office of Science to advance the research and infrastructure priorities of the Advanced Scientific Computing Research (ASCR) program.
Upcoming Funding Opportunities
Within the next two months, ASCR plans to release four FY 2019 funding opportunity announcements open to research universities and national laboratories:
- Quantum algorithms. ASCR would like to fund additional quantum algorithm teams to build on the FY 2017 awards. DOE will be seeking basic research proposals that exploit advances in quantum simulation and machine learning algorithms that focus on key topics of most relevance to the Office of Science, such as quantum simulation of quantum field theories, hybrid and multiscale simulations, simulation of non-equilibrium dynamics and thermodynamics, and quantum machine learning algorithms that combine methods from quantum information science (QIS) and machine learning. DOE anticipates up to three awards at $1 million a year over three years.
- Quantum networking. This is a new area of investment for ASCR focused on early investments to develop scalable and adaptable quantum network infrastructures that can support the transmission of diverse types of quantum information. One of the main goals is to deploy new quantum networks that can co-exist with DOE’s existing Energy Science Network (ESnet), which allows the national labs and major research collaborators to share scientific information and resources over 13,000 miles of coast-to-coast dedicated optical fiber. The first funding call is likely to focus on novel quantum network architectures, including photonic quantum networks and optical fiber systems and quantum network devices and subsystems, including transduction devices, quantum repeaters and routers, and quantum frequency conversion. DOE recently completed a workshop report on “Quantum Networks for Open Science” which provides more details on future research directions and is available here.
- Scientific machine learning and uncertainty quantification. The focus of this funding call will be on developing uncertainty quantification methods for scientific machine learning applications that take into account the reliability and usability of data that are noisy and uncertain and often incomplete, sparse, and only partly informative. There is interested in using machine learning for discovering correlations in data sets and uncertainty quantification can add significant robustness and realism to those applications.
- Co-design center for artificial intelligence, machine learning, and data analytics. DOE Co-Design centers are partnerships between national laboratories and research universities to develop hardware, software, and algorithm solutions to help solve specific scientific problems. In this case, a team of vendors, hardware architects, system software developers, domain scientists, computer scientists, and applied mathematicians would work together to design and understand the various tradeoffs of artificial intelligence and machine learning applications on system architectures, hardware, software, and algorithms.
Funding Outlook for ASCR
In FY 2019, Congress appropriated $936 million to this program, an increase of $289 million or 45 percent above FY 2018. The increase was primarily driven by the needs of the exascale computing initiative which aims to deploy the first two exascale computing systems at Argonne and Oak Ridge National Laboratories in 2021 and 2022, respectively. Additional funding was also provided to expand research efforts in QIS and artificial intelligence and is reflected in the funding opportunities listed above.
The FY 2020 President’s budget request for ASCR is $921 million, a decrease of $15 million or two percent below FY 2019. The funding decrease is primarily driven by the end of peak funding for the exascale computing project and the program is planning to use carry over funding reserved in prior years as contingency for the exascale project to pay for facility upgrades at Argonne and Oak Ridge National Laboratories. The primary focus in FY 2020 is to grow applied mathematics and computer science research funding to support artificial intelligence, machine learning, QIS, and other beyond Moore’s law technologies. The FY 2020 budget request proposes a $16 million or 12 percent increase to applied mathematics and computational science research. The table below provides detailed budget information. To further grow fundamental research, ASCAC formed a Subcommittee on Transition from the Exascale Project to provide advice on new research priorities for applied mathematics and computer science, chaired by Dr. Roscoe Giles from Boston University, that will restore and shift funding from the exascale project into fundamental research, maintain the software, heardware, and applications for exascale systems, and further develop the workforce. This subcommittee plans to release a draft report in September that will guide future budget discussions.
Future Research Priorities and FY 2020 Funding Opportunities
The two main research priorities over the next 18 months are:
- Artificial intelligence and machine learning. The FY 2020 President’s budget request proposes $36 million for artificial intelligence within ASCR, an increase of $23 million or 177 percent above FY 2019. The focus is on addressing basic research needs in scientific machine learning and extremely heterogeneous systems. The major funding opportunities in FY 2020 are likely to be:
- Foundational research in applied mathematics for scientific machine learning. The focus is on improving the reliability, robustness and interpretability of big data and artificial intelligence technologies and start to develop new algorithms, methods, and software tools for extracting information from scientific and engineering data. DOE recently released a report on Basic Research Needs for Scientific Machine Learning: Core Technologies for Artificial Intelligence, which is available here. The contents of this report will form the basis for DOE’s future investments in scientific machine learning. DOE is particularly interested in using machine learning to derive scientific discovery from the data generated by its user facilities and the exascale computing systems that will soon be deployed. In order to realize this vision, the report identifies six Priority Research Directions which are equally divided between two themes: foundational research, which corresponds to domain awareness, interpretability, and robustness; and capability research, which focuses on data analysis, machine learning-enhanced modeling and simulation, and intelligent and automated decision-making. The table below summarizes future research priorities and directions.
- Co-Design Center for a Distributed Computing Ecosystem. A new Co-Design center that would develop hardware, software, and algorithms needed to integrate big data to support the large-scale computing and data requirements for machine learning.
- QIS. The FY 2020 President’s budget request proposes $51 million for QIS within ASCR, an increase of $17 million or 50 percent above F 2019. The major funding opportunities in FY 2020 are likely to be:
- Quantum networking. Another funding call to build on FY 2019 awards on quantum networking, as discussed previously.
- Quantum science and technology center. The FY 2020 President’s budget request provides funding for at least one quantum science and technology center authorized by the National Quantum Initiative Act. DOE is hoping that Congress will provide additional funding to support up to five centers, but currently funding is included for only one center. These five-year, multi-disciplinary centers with funding of up to $25 million a year would focus on addressing scientific grand challenges related to advancing quantum applications in quantum computing, sensing, networking, and communications.
With the exascale computing project nearing completion and the end of Moore’s Law quickly approaching, DOE had charged ASCAC with identifying opportunities and challenges for future high performance computing capabilities and recommending areas of research and emerging technologies that need to be given priority in the future. ASCAC released its report on Future High Performance Capabilities on March 20 and it is now available here.
- reconfigurable logic,
- memory-centric processing,
- silicon photonics,
- neuromorphic computing,
- quantum computing, and
- analog computing.
The Advisory Committee emphasized that there will be a period of uncertainty over the next decade on the future trajectory of high performance computing and a number of approaches may be highly disruptive. In addition, the future of computing in the post-exascale and post-Moore eras will be defined by extreme heterogeneity. The challenges and opportunities in an era of extreme heterogeneity are highlighted in a recent report available here. One of the main recommendations to DOE is greater investments in applied math and computer science to be ready for this new era of computing and recruiting, growing, and retaining a future workforce.