Remarks

Thank you for that kind introduction, Dr. Luebke. It is truly a pleasure to be back here again at the NVIDIA Developers Conference. I'm thrilled to help introduce this panel on fundamental research, since that's what we at the National Science Foundation are all about.

[Introducing Pasteur's Quadrant]

When we think about the role of fundamental research and where it takes place in our society, I'm always reminded of the two-dimensional classification scheme that Donald Stokes put forward in his 1997 book, Pasteur's Quadrant. Stokes divides research along two dimensions -- one dimension describes the quest for fundamental understanding; the second captures use-inspired pursuits. These two dimensions divide the research space into four quadrants, three of which are relevant here:

• curiosity-driven fundamental research, exemplified by the physicist Niels Bohr -- or today's LIGO pioneers who are receiving this year's Nobel prize in physics for the discovery of gravitational waves;
  
  
• use-inspired basic research, exemplified by the biologist Louis Pasteur -- and by today's pioneers in gene-editing; and
  
  
• pure applied research, exemplified by the great American inventor Thomas Edison -- and today's inventors of the Internet and the World Wide Web.
  
  

In thinking about these two dimensions to research, I sometimes rue the lack of a third dimension -- that of time. Individual researchers, by their very nature, may prefer a particular research style that falls within one quadrant. But when we think of an idea and its evolution -- its genesis ... its development ... and all that it may blossom and follow in its wake -- we see the flows and eddies of that idea over time among all three quadrants.

[Machine Learning and the flow among quadrants]

The reasons that we are all here today -- deep learning algorithms, the petaflop computational power harnessed in NVIDIA's GPU ("G-P-U") systems, and the application of deep learning in myriad endeavors -- these touch all three quadrants.

The idea of Artificial Neural Networks began as curiosity-driven fundamental research, inspired by biological neural networks in early brain research. The curiosity-driven question -- what might such simple circuits be able to compute? A simple neural net machine, the Perceptron, was built by Frank Rosenblatt as early as 1957. Researchers then began to explore neural networks as classifiers -- use-inspired research enabled by, and flowing from, earlier curiosity-driven research. Neural net research went quiescent as other techniques achieved better classification performance.

But then in 2006, breakthrough work by Geoffrey Hinton enabled tremendous efficiencies in training multi-layer neural networks. Dr. Hinton was an NSF-funded faculty member at Carnegie Mellon University before he moved to Toronto.

Computation remained a limiting factor until computing power unleashed through GPUs led to today's "deep learning" revolution. This part-curiosity-driven, part-use-inspired idea of exploiting GPUs -- processors designed for graphics and gaming -- in supercomputing and machine learning applications is due in no small part to our moderator, David Luebke. Dr. Luebke was a successful NSF-funded researcher at the University of Virginia before he joined NVIDIA.

Today, deep learning has enabled computers to achieve human-capability levels in interpreting images, speech, and language. Deep learning research abounds in the use-inspired and applied research quadrants in both academia and industry, and plays a key role in our Nation's economic competiveness and national security. It is powering automated control systems, winning strategies for board games, and science at scales as large as the universe -- the search for distant stars and galaxies -- and as small as the atom -- predicting molecular interactions that cause disease.

Looking over time then, we see the flows and eddies of an idea that began as curiosity-driven research but has given rise to innovations and industries in the use-inspired and applied research quadrants.

Moreover, these flows and eddies can return back to curiosity-driven fundamental research. Can deep learning be a model not just for lower-level perceptual AI tasks, such as speech recognition, but also for human-like reasoning and decision making by machines? What can a neural network compute? Over time, a mix of curiosity-driven and use-inspired research will tell us.

[The American IT Innovation Ecosystem -- A Flow of Individuals, Ideas, and Innovations]

The mission of the National Science Foundation is "to promote the progress of science; to advance the national health, prosperity, and welfare; to secure the national defense." We make sustained, long-term investments in discoveries and discoverers in both curiosity-driven and use-inspired fundamental research. And these individuals and ideas give rise to further innovations in pure applied research led by industry.

These flows of ideas -- and of people and artifacts -- among quadrants and among academia and industry in computing has been characterized by the so-called "tire-tracks" diagram published by the National Academies. [Note: NVIDIA is on the tiretrack diagram below.] This flow of ideas over time has been critical in establishing and fueling many of the major information technology industries in our Nation.



Slide 2 image: graph showing the contributions of federally supported fundamental research to the creation of IT sectors, firms and products with large economic impact from 1965 to 2010.

Credit: Reprinted from National Research Council, 2012, Continuing Innovation in Information Technology, The National Academies Press, Washington, D.C.

[Partnering With You]

We at NSF are on a grand mission to invest in discoveries and discoverers across the Nation. In the last year, we identified 10 Big Ideas for Future Investment -- bold foward-looking ideas that NSF is uniquely positioned to address. One of these, the Future of Work at the Human-Technology Frontier, is focused on how advances in AI, machine learning and other technologies -- advances that we will hear about here today -- will impact the future of work.

We, like you in industry, are part of a great research ecosystem that has been an economic engine for our country for decades. Partnerships among academia, industry, and the Federal government will continue to be critical to accelerating advances in our field; the tiretracks diagram illustrates one aspect of these partnerships.

We at NSF are working to build new kinds of partnerships in which industry directly co-invests with us in our research funding. Our Directorate for Computer and Information Science and Engineering has led several such collaborations -- and we are growing new ones. We're doing so because we believe we offer a unique convening power to forge such partnerships -- and it is only through these that we can fully leverage the unique strengths, capabilities, and resources that we each bring to our Nation's research ecosystem.

While our overall budget at NSF has been flat during this decade, through imagination, resourcefulness, and above all -- partnerships -- our ability to do the fundamental research that leads to invention will continue to grow and return dividends to this country.

[Closing]

I look forward to hearing about research advances here today in the complex flow and eddy of discovery and invention, as ideas advance from basic to applied, and as applications emerge and more questions are uncovered, leading back to basic research. Even more, I look forward to the new partnerships that will form and current partnerships that will deepen here at this Conference and well into the future. I am excited to work together with all of you as we pursue our grand mission. Thank you again.