Over the past three years, Andrew Ng, an associate professor of computer science, has noted a paradigm shift in the role of tech companies on campus. Specifically, Ng sees what he calls a “new model of research,” which he believes exists in many disciplines, but is especially prominent in his own field, computer science.
Previously, Ng had conducted his research along the traditional “tech transfer” model.
“You invent something, write it up and then hopefully someone tries to build a product or look at the technology and find a use for it,” Ng said. “That’s what most people, including me, were doing maybe 10, 15 years ago.”
Although Ng said this traditional approach is by no means obsolete, he has watched computer science research shift increasingly toward the more product-driven, immediate-impact-oriented research model of corporations and corporate labs — a shift he speaks about with great enthusiasm.
“I’m seeing a trend of research where researchers, either in corporations or in academia, are trying to get a better understanding of what advances in technology are most likely to help people’s lives get better,” he said.
Ng is not alone in his observations. Multiple Stanford professors point to the growing role of industry in computer science research, and the implications — both positive and negative — of this “new model.”
Shifting funding landscape
Professors attribute the shift toward industry to a decline in traditional government funding for research, as well as the need for increasingly large resource pools to support complex projects.
Bill Dally M.S. ’81, professor of computer science and electrical engineering, has seen a significant decrease in government support since he began teaching in the 1980s.
At the time, he said, the main bankrollers of CS research, including DARPA (Defense Advanced Research Projects Agency), the NSF (National Science Foundation) and DOE (Department of Energy), were more generous and more willing to back wilder ideas.
“It was pretty easy for people to go and work on very far-reaching ideas without any immediate practical application,” Dally said. “There was a lot of support for that.”
Dally and John Ousterhout, a professor of computer science and electrical engineering, both believe that this relatively unrestricted funding for tech research enabled some of the most important developments in computer science.
The Internet grew out of a DARPA project begun in 1973; Larry Page and Sergey Brin developed PageRank, the algorithm that led to Google Search Engine, with the help of $4.5 million from the NSF.
But over the past 10 or 20 years, according to Ousterhout and Dally, government funding has become harder and harder for researchers to procure. For example, DARPA’s grants today are smaller and place more strictures on how researchers can use their funding.
“All of the large programs — MIT, Berkeley, Stanford, Carnegie Mellon — all suffered because they don’t have these chunks of money that allow us to the do the most innovative and exciting research,” Ousterhout said.
At the same time, ever-more-sophisticated technology demands resources that, oftentimes, only industry can provide. As a result, said Ng, “a lot of research that used to be done in academia now is done in corporations.”
“In the world of databases and the world of [artificial intelligence], where I am, it’s increasingly difficult for just academia to have the resources to do things at that scale and to build the giant databases or even have access to the data,” he added.
Ng also cited microprocessors, products of academic innovation that he said are now so expensive to make that they are developed almost exclusively within companies like Intel and NVIDIA.
In search of funding
These funding changes have encouraged Dally and Ousterhout to shift more of their time and research into an industrial setting.
Dally has gone from being a full-time faculty member at Stanford to spending one day a week at the University, in order to devote more of his time to NVIDIA, where he is senior vice president of research.
“I felt that if I was going to be limited on getting funding to do industrial research, I might as well do it in industry, where I have much better resources and can get a lot more work done,” Dally said.
Ousterhout had similar problems obtaining non-industrial funding for his research. Since his return from industry to academia, he has submitted four NSF grant proposals, only one of which was accepted. The project he believed to be the most exciting of the four, Ram Cloud — a long-term project aimed at building a storage system thousands of times faster than existing ones — he submitted twice, only to be rejected both times.
After failing to receive government funding, Ousterhout was able to receive industrial funding, and he has been working on Ram Cloud for the past five years.
These professors’ experiences exemplify a broader funding trend. While the majority of external funding for computer science research at Stanford still comes from the federal government, industrial support has risen rapidly in past few years.
In 2014, the latest year for which the American Society for Engineering Education (ASEE) provides data, approximately 68 percent of the $25 million total external funding for Stanford computer science research came from government. Twenty-two percent came from industry — a significant increase from just one year earlier, when industry contributed 14 percent of the total, and from 10 years ago, when industry contributed just 3.8 percent.
And today presents a very different picture from 2000, when industry accounted for exactly zero percent of research funding — not only in computer science, but across most engineering disciplines.
Potential for impact
Many Stanford professors recognized the shift toward industrial research as beneficial to both researchers and users of the technology that results — another reason why professors are drawn to industry.
Since 2011, Ng has divided his time between Stanford and companies including Google, Coursera and now Baidu, where Ng is chief scientist.
“One of the things at Baidu is it gives researchers access to the data, the problems that touch people’s lives,” Ng said.
Ng cited a particular Baidu project in China as an example of industry’s ability to strategically target research toward tangible impact and users’ needs. Ng worked on a wearable camera for blind individuals that can describe images of surroundings to the wearer.
The practical needs of blind users in China helped Ng and his colleagues prioritize their research directions. For example, blind users there requested a tool that would help them recognize different denominations of money, as the similar size and natural wear of bills’ texture resulted in money being indistinguishable by touch.
“Feedback from the people we were trying to help, blind people, was really crucial in helping us realize what was important to work on,” said Ng.
Similarly, Subhasish Mitra Ph.D. ’00, associate professor of electrical engineering and computer science, enjoyed the large impact he experienced while working at Intel.
“By the time I was leaving the company, there were close to 50 product teams using the technology that I created,” Mitra said. “Today, almost every digital computing system uses what I created.”
Mehran Sahami ’92 M.S. ’93 Ph.D. ’99, professor of computer science, saw similar advantages of industrial research.
“When I was at Google… if you come up with some sort of interesting result and push it out, you’re impacting the search engine that’s getting billions of queries a day,” Sahami said. “That’s just a scale of research that’s larger than what you get at a university.”
Both Mitra and Sahami, however, eventually returned to academia, leaving their industry positions to work with students.
Downside of short-term focus
Some researchers worried that the short-term, practical focus of industrial research can neglect important longer-term discoveries.
Dally is particularly interested in improving “parallel programming,” or making it easier to program many processors to work together. However, he says companies are unlikely to support this line of research because it will not pay off for five or 10 years; they prefer projects that yield practical results in two or three.
“I think this is very unfortunate because it leaves this big gap of stuff that’s not getting done that we actually do need for the future,” Dally said.
While Mitra works closely with industry for much of his research, working in academia has allowed him to pursue more radical projects. Right now, he is working on “N3XT,” which leverages carbon nanotubes and new memory technologies to hopefully achieve “the next thousand x” of performance for computer systems.
“In that domain, the industry actually wants academia to lead the way through revolutionary ideas and foundational work before they pick it up,” Mitra said.
Ousterhout lamented government’s decreased willingness to complement industrial funding by enabling the sort of long-timeframe research in which industry has less interest.
“Ultimately, I wonder if a few of the top schools like Stanford will figure out ways to increase their endowments so they can endow not just their faculty’s salaries but also research programs,” Ousterhout said.
He admitted that it would be impractical today to replace all of the research money that enters campus. Stanford states that it hosts $1.22 billion in externally sponsored research projects. Of 5,500 projects, 81 percent are funded by the government.
Nonetheless, Ousterhout believes it is important to give researchers freedom of exploration.
“It might sound sort of irresponsible to just throw money at people, but that’s actually the way research works best,” he said.
As researchers consider the benefits and drawbacks of both university and industrial settings, computer science students navigate a related question: Should they pursue work at a company or in the academic world?
Curious about the pull of industry versus academia for students, The Daily surveyed via email 130 Stanford undergraduate and graduate students interested in computer science. About one-third of respondents had participated in academic CS research, while three-fourths had worked or interned in CS industry.
When asked about their path or intended path after finishing their undergraduate years, a clear majority of students — 65 percent — said they chose or would choose to work in industry. About 15 percent chose or would choose academic work, while another 16 percent were unsure and 4 percent selected “Other.”
Many of those drawn primarily to industry — both as a summer activity and an eventual career — echoed professors’ desire to maximize the practical and immediate impact of their work on others’ lives.
“My primary goal within computer science is to leverage technology to create social impact,” wrote one sophomore. “There are definitely avenues of research which tend towards this goal, but the returns are much longer-term. I see huge potential for addressing more immediate problems and, perhaps more importantly, changing the culture and focus of the tech industry to center social impact alongside innovation and profit.”
However, the same sophomore also admitted that other factors influenced her choice to work in industry over the summer.
“There’s a pressure to land ‘prestigious’ internships, and a sense that research is second-tier or for people who couldn’t get another job,” she wrote.
One junior explained over the phone that she felt a “push at Stanford and especially in the computer science department” to choose industry. The pressure she described was indirect but pervasive, the inevitable consequence of being surrounded by company visits to campus, emails advertising tech jobs and buzz about startups.
She also noted that tech companies recruit primarily during fall quarter while research positions recruit mostly in winter; it is difficult to turn down a fall job to consider other options, she said.
But for this student and many others, industry’s biggest advantage over academic research is simple: more money.
A job at a tech company usually not only pays more but comes with perks such as free housing and food. In contrast, the junior explained, a campus research position would give her just enough funds to cover living and some travel expenses.
Although she is interested in academic research, an industry job means she won’t have to ask her parents to pay for her plane tickets home. It means she can cover her expected tuition contribution and not work during the school year.
“Personally I have a lot of friends who also face this dilemma who are in CS,” the junior said. “They all go through this and also choose tech companies because they want to send money back home.”
Stanford’s CURIS program, which connects CS students with summer research projects on campus, raised its stipend this year to $6,400 for 10 40-hour weeks. However, this is still below what many tech companies offer.
According to Glassdoor, Google and Facebook interns earn an average of $7,120 and $6,294 per month, respectively; a summer intern could earn over $20,000. Twitter, Apple and NVIDIA interns earn between $30 and $40 per hour, while CURIS’ pay works out to $16 per hour.
Some students reported turning down CURIS positions for financial reasons in the past — one senior chose an internship at Symantec over a security-related CURIS project for the higher pay, although in retrospect, he believes he would have enjoyed CURIS more.
Students who chose academic research over industry said they appreciated the greater flexibility and freedom that academic work affords. For Raghu Prabhakar, a third-year Ph.D. student in computer science with research experience both in industry and on campus, his work at Stanford has been more rewarding.
“You might earn a bit more by going to industry, but if what you’re doing in the industry is not directly related to your project or your research, then you’re wasting your time in the long run,” he added.
Sahami said that students’ attraction to industry “comes and goes on campus.” After the burst of the dot com bubble in 2000, he said, interest in startups and entrepreneurship lessened; right now, interest is on the rise again.
“At the end of the day, it’s about what students want to do,” Sahami said. “They should do what they’re really passionate about.”