NAIRR Pilot Presents Opportunity for Statisticians

To strengthen the connection between the statistical community and National Science Foundation, we continue the series introduced in the May 2023 issue that poses questions to NSF program officers and awardees. If you have questions or comments for the program officers, send them to ASA Director of Science Policy Steve Pierson.

This month’s Q&A spotlights NSF’s National Artificial Intelligence Research Resource (NAIRR) pilot that the foundation launched in January and the Office of Advanced Cyberinfrastructure within the NSF Directorate for Computer and Information Science and Engineering.

Shrijita Bhattacharya is an assistant professor in the department of statistics and probability at Michigan State University. She earned her PhD in statistics from the University of Michigan, Ann Arbor, in 2018. Her research interests focus on variational inference–aided scalable Bayesian machine learning with applications to neural networks, Ising models, and computer models. She is the recipient of the NAIRR pilot grant and an NSF-collaborative research grant.

Please summarize the project and what it will accomplish.

The grant aims to develop a scalable Bayesian approach for learning the lower dimensional manifolds of large data sets, especially if the lower dimensional manifold has a convoluted nonlinear structure. Existing statistical approaches suffer from scalability issues with an increase in the sample size or dimension of the feature space. Our proposed Bayesian solution is not only scalable but also captures the best nonlinear manifold at a given intrinsic dimension, along with the uncertainty associated with the dimension itself. This opens the opportunity to save large data sets on low latency devices such as mobile phones, tablets, and low-memory computers for reuse on user-specific downstream tasks.

Since training on large data sets requires huge computational resources, the NAIRR pilot presents an exciting avenue for statisticians entering the artificial intelligence field. The NAIRR pilot is an initiative led by the National Science Foundation in collaboration with other federal agency partners and nongovernment partners.

If an NSF entity other than the Division of Mathematical Sciences partially or fully funded the award, please describe your approach to that entity.

Since the NAIRR pilot is a request for computational resources, the project team investigated the resources provided by each of the participating agencies and carefully designed the layout of the amount and time of resources. View the distribution of the available resources from each participating agency.

What advice do you have for others applying for NSF funding?

The NAIRR pilot proposal requires the submission of a three-page proposal. There are four components to this proposal scientific/technical goal, which includes motivation of the problem, its technical challenges, and its broader impact; estimation of computational resources with a description of the split between GPU, CPU, and memory requirements; support needs, which describes the help needed from support staff maintaining the resources; and team preparedness, which includes any preliminary work and detailed plan for using the allocated resources.

How do you envision the challenges and opportunities of the statistical community in this era of artificial intelligence?

Interpretable artificial intelligence requires a solid statistical and mathematical foundation of any proposed machine learning model. The statistical community can contribute to this era of interpretable AI if they can establish the scalability of the traditional statistical approaches to humongous data sets. However, storage and implementation of proposed algorithms on these data sets impose a heavy computation burden.

Without computational resources at one’s disposal, the statistical community will face challenges in adapting to this modern era, in which tons of data are pouring in every second. The NAIRR pilot presents a most-needed opportunity for statisticians to contribute to the world of statistically principled AI.

_____________________________________________________________________________________

Wen-wen Tung is on leave from Purdue University, where she served as a member of the teaching academy and was the founding director of the geodata science for professionals master’s program from 2019–2024. She is now in her first year as a rotator program director in the Office of Advanced Cyberinfrastructure within the NSF Directorate for Computer and Information Science and Engineering.

What are the programs of the Office of Advanced Cyberinfrastructure most relevant to the statistical and data science community, and what are their scopes?

The Office of Advanced Cyberinfrastructure, referred to here as OAC, supports the development of an integrated ecosystem of cyberinfrastructure resources to accelerate research and innovation. This includes advanced computing, data and software infrastructure, networking capabilities, centers of excellence, user-facing services, and initiatives in learning and workforce development.

Computing environments have long been ingrained in the technical fields of data science, as elucidated by William Cleveland in the International Statistical Review article “Data Science: An Action Plan for Expanding the Technical Areas of the Field of Statistics” and David Donoho in his Journal of Computational and Graphical Statistics article titled “50 Years of Data Science.” OAC programs are more relevant than ever to the statistics and data science communities, especially now as big data, AI, and digital twins drive unprecedented demand for advanced cyberinfrastructure. Through OAC, NSF leads several focus areas in the National Artificial Intelligence Research Resource (NAIRR) Pilot, which envisions a shared national research infrastructure for responsible discovery and innovation in AI. The National Discovery Cloud for Climate, another OAC-led initiative, enables access to advanced computing, data, software, and networking resources to support climate research and its broader applications.

Moreover, researchers and educators can apply for NAIRR Pilot resources, regardless of whether they have NSF grants. Similarly, advanced computing users can apply for resources through the NSF ACCESS program. Overlooking these resources when developing a project proposal requiring advanced computing would be a missed opportunity.

What kind of research proposals does OAC fund out of its Core program?

The OAC Core program supports research that enables future cyberinfrastructure. However, unlike other core programs in the Directorate for Computer and Information Science and Engineering, OAC Core only funds small projects (up to $600,000 total budget). This is because OAC funds several significant regular cyberinfrastructure programs that complete an integrated ecosystem: Cyberinfrastructure for Sustained Scientific Innovation, responding to the evolving and emerging needs in cyberinfrastructure; CyberTraining, providing training on cyberinfrastructure for research; Strengthening the Cyberinfrastructure Professionals Ecosystem, offering training to cyberinfrastructure professionals; and Cybersecurity Innovation for Cyberinfrastructure, supporting research on securing scientific data, workflows, and infrastructure.

Does the OAC program support PIs whose primary appointment is in a statistics department?

The integration of statistics and advanced computing with data presents immense opportunity and empowerment. Statistics PIs are encouraged to explore the full range of OAC funding programs. Furthermore, OAC-funded projects are characteristically multidisciplinary and emphasize broader impacts.

Projects with a strong scientific or engineering driver that deliver significant societal benefits align particularly well with OAC’s values. PIs should also consider OAC co-funding opportunities such as the Computational and Data-Enabled Science and Engineering and Accelerating Computing-Enabled Scientific Discovery.

Finally, collaborating with advanced cyberinfrastructure professionals can significantly enhance the innovative use of cyberinfrastructure in a proposed project, thereby increasing the likelihood of funding success.