How an innovative grants program (and Belgian beer mixers) at Johns Hopkins fuels discoveries about the human brain
A neuroscientist, an electrical engineer, a surgeon, and an education researcher walk up to a bar.
This could be the start of a joke, or it could be a scene from a recent Science of Learning Institute event at Johns Hopkins University. At the institute's four-times-yearly Belgian Beer Events, scientists from far-flung fields—and often from far-flung parts of the university itself—present their research to each other in short, digestible chunks. Their creativity and conviviality stimulated by a cup of ale or lager, the researchers strike up conversations and form connections that range widely across disciplinary boundaries, from classroom learning to machine learning, from recovery from stroke to memory formation in the brain.
Such conversations can be all too rare at a university where faculty are spread not just across a campus but throughout a large city and beyond. The result, for an inherently interdisciplinary subject like the science of learning, is that projects that could address fundamental and important questions can be hard to conceive and get off the ground. And too often, promising basic research doesn't get translated into the settings where it could help real-world learners.
The Belgian Beer Events, conceived shortly after the institute launched in 2013, are helping change that. They provide an informal space where basic researchers can meet translators, where machine-learning experts can meet early-childhood educators, where cognitive scientists can meet smartphone app developers. The events rotate between locations: October's was at the School of Education, and December's was hosted by the Department of Biomedical Engineering; previous ones were held at the School of Medicine and in Homewood's Levering Hall. Computer scientist Greg Hager likens the events to "an intellectual mixing bowl."
Beyond generating lively conversation, the gatherings are sparking collaborations between researchers who otherwise might never have met. At an event in 2013, neurologist Bonnie Nozari presented her work on speech and language processing disorders. Computer scientist Raman Arora then spoke about his work on machine learning and speech recognition. Recognizing a mutual interest in speech, the two chatted. The next day, they began planning a joint project to see if computers can predict how humans will pronounce words, and then provide feedback to people seeking to learn a new language, or to relearn how to speak after a stroke.
It sounds like a lucky encounter, but in fact electrical engineer Sanjeev Khudanpur, a member of the institute's steering committee, was at work behind the scenes. He conceived the Belgian Beer Events, and he made sure that Arora, his colleague in the Whiting School of Engineering, would be speaking on the same day as Nozari, of the School of Medicine. Later, when the two were ready to apply for funding, Khudanpur encouraged their ultimately successful proposal for one of the institute's research grants. "I see myself as a matchmaker," he says.
"It's that kind of really innovative, different seeding of projects that I think we've done really well," says Barbara Landau, the institute's director and the Dick and Lydia Todd Professor of Cognitive Science in the Krieger School of Arts and Sciences. The institute funded eight projects in 2013 and eight more in 2014, with projects receiving an average of $140,000 spread over two years. Funding goes to hiring graduate students and postdoctoral researchers, developing software, purchasing equipment, and supplying other research needs. The grants are competitive; the review committee has received around 30 proposals a year. The funded projects address a broad range of learning settings, from the classroom to the operating room to distance learning that can take place anywhere. The learners are not limited to humans, either; many of the projects include a strong component of "machine learning"—harnessing computers to recognize patterns in data and use them to develop new human learning applications. Other projects focus on developing animal models that can be used to study human learning.
The grant program allows researchers to get support for projects that might not be quite ready for a proposal to a traditional funding agency like the National Science Foundation or the National Institutes of Health, says Landau. Almost without exception, an NSF or NIH review panel will want to see at least preliminary data demonstrating that an idea is viable. With Science of Learning Institute funding, scientists can do exploratory research that will provide the data needed to support a larger proposal to a more traditional funding agency. "It allows people to do things that they wouldn't necessarily be able to accomplish by a standard grant," says Landau. "The granting agencies tend to be somewhat conservative, and we're looking for innovation."
Like Arora and Nozari's collaboration, many of the funded projects harness existing technological applications to improve learning, often in novel ways. For example, Khudanpur and Hager are working with Gyusung Lee, an instructor of surgery in the School of Medicine, to develop computer software that can help teach surgeons how to use the da Vinci robotic surgical platform. The project grew out of an existing effort called the Language of Surgery, developed by researchers in the Whiting School of Engineering's Laboratory for Computational Sensing and Robotics.
Through this effort, which began in 2006, Hager, Khudanpur, and colleagues program computers to record and analyze the different kinds of movements that surgeons make while performing certain tasks with surgical robots. The researchers' goal was to find movements that could consistently be classified as either expertlike or novicelike. Novice surgeons are more likely to break a suture, for example, or to push or pull on tissue while using the robot to manipulate a surgical needle. The researchers were able to train computer software to recognize such expert and novice movements much as a surgical trainer would.
The next step is to have the assessment tool provide real-time feedback to surgical trainees. With the kind of application the researchers are envisioning, trainees could, in theory, receive an unlimited amount of individualized feedback on what skills they have mastered and where more work is needed. "We're putting the computer in the human learning loop," Khudanpur says. "The computer has certain abilities that are complementary to humans. [For example,] the computer doesn't get tired. The computer usually doesn't charge by the hour."
A few years ago, when the researchers applied for an NIH grant to develop such a learning application, the proposal was rejected because they had no data showing the idea had promise. Thanks to their Science of Learning Institute research award, the scientists are starting to collect that data. Backed by some preliminary results, they recently put in a new NIH proposal and are waiting to hear back.
Meanwhile, thanks to a talk Hager gave last fall, his team's research may soon spawn another effort, which would take Language of Surgery technology out of the operating room and into the classroom. Hager's presentation inspired Landau and Amy Shelton, a professor in the School of Education, who is also on the institute's steering committee, to wonder whether motion-tracking software could recognize the movements that young children make when learning to build toy towers out of blocks. Spatial skills like tower building, in addition to being important in their own right, are of interest to researchers because they often predict children's future abilities in math and other areas. Hager, Landau, and Shelton are now discussing a potential project to put motion sensors on blocks and use computers to track how children acquire manipulation skills, a tactic similar to the one Hager's team uses to assess the skills of aspiring surgeons.
Institute-funded collaborations between computer scientists and education researchers are also reaching far beyond traditional education settings like medical training. In a project funded in 2014, computer scientists Philipp Koehn and Jason Eisner are teaming with Chadia Abras in the School of Education's Center for Technology Education to develop a radically new way to learn a foreign language. The idea is based on macaronic language—a kind of text that mixes two languages into a Spanglish-like hybrid. While such mixing has traditionally been employed by novice speakers or for satirical purposes, Eisner realized that coupled with recent advances in machine translation, it could also help introduce learners to foreign vocabulary and syntax in a gentle and piecemeal way rather than all at once, as in a typical foreign text read laboriously with the aid of a dictionary.
To implement the idea, the researchers are developing software that translates a text progressively, with more and more of the text appearing in the foreign language as the reader's comprehension improves. For an English-to-German learner, for instance, the English phrase "a loaf of bread" could start to appear as "ein Loaf of Bread." When the reader is comfortable with reading the German word "ein" instead of the English "a," the program could progress to "ein Breadloaf," resembling German in syntax but retaining English words. The text would then become "ein Brot loaf," and finally the fully German "ein Brotlaib*." The program will intermittently assess the student's reading comprehension and ability, and tune the amount of foreign language presented to the reader's progress; readers also can direct the program to make the translation easier or harder.
Since the concept still needs to be proved, it makes an ideal Science of Learning Institute project, says Koehn. Eisner adds, "It's a bet that this will work out and will not, for example, confuse people or give them bad habits." The researchers plan to develop an English-to-German application and test it on the Web and in Johns Hopkins classes in combination with more traditional classroom and textbook instruction. If successful, the software could also be made available on the Internet for independent learners.
The project exemplifies how interdisciplinary teams can merge cutting-edge research in machine and human learning, says Kelly Fisher, the institute's assistant director and an assistant professor in the School of Education. "It's a software program that is learning itself, learning about the learner."
Institute-funded research also targets learners far beyond those who are acquiring skills for the first time. Learning is critical for the millions of people who lose skills when they suffer strokes and other neurological conditions and then need to regain them, often through lengthy and complex rehabilitation processes. Research on how to more efficiently relearn lost skills could make a huge difference in how quickly such people can return to work and fully participate in society again.
Cognitive scientist Michael McCloskey recently discovered a new, debilitating, and apparently very rare reading deficit known as alphanumeric visual awareness deficit, or AVAD. McCloskey, a professor in Cognitive Science, identified the condition based on two cases that came to him in one year. One of them, a 61-year-old Baltimore geologist with a neurological disease, could see fine in general, but when looking at letters or numbers, he saw only blurs. McCloskey and his colleagues found, however, that by teaching the patient new characters to use in place of the digits, they could restore his recognition abilities. The researchers developed a smartphone calculator app and modified the geologist's laptop to allow him to do math with the new symbols.
Seeking to build on this work, McCloskey assembled a team of neurologists and cognitive scientists to look for more people with AVAD in order to study the condition using brain imaging and other techniques, and to develop apps and other technology that would help affected people make sense of letters and numbers again. But the researchers have run into a roadblock: They haven't found a single other case of AVAD beyond the original two. A woman in North Carolina who seemed to have the deficit turned out to have a somewhat different condition. "On the one hand, it's interesting that [AVAD is] so rare; on the other hand, it's not what we were hoping for," McCloskey says.
So he and his team have reoriented their project, broadening the scope to include more-common character recognition disorders. For example, some people cannot recognize a number or letter when it is presented to them whole but can recognize a character if they watch it being drawn. Perhaps, says McCloskey, a smartphone app could be developed to read signs and other important text, and draw each character in sequence for people with this deficit. His team is also starting to collaborate with a software developer, MicroBLINK, to make an app that would identify characters and then read the text aloud.
In addition to potentially helping people regain lost abilities, many institute-funded projects such as McCloskey's are aimed at teasing apart the different brain regions and processes responsible for seemingly coherent learned skills like reading. Along these lines but focusing on an entirely different brain function, psychologist Marina Bedny, of the Krieger School's Department of Psychological and Brain Sciences, is heading a team that received an institute grant to study how the brain can retool its hardware when the original purpose of one of its regions is no longer needed. In sighted people, around a quarter of the brain is devoted to visual processing; in blind people, these brain regions get repurposed. How does this work? Bedny wondered.
To investigate this question, she and colleagues in the Krieger School's Department of Cognitive Science and in the Department of Physical Medicine and Rehabilitation at the School of Medicine are combining language comprehension assessments with a technique called transcranial magnetic stimulation, or TMS. They hope to learn whether brain regions normally devoted to sight are needed for language processing in blind people. The researchers recently collected data at a National Federation of the Blind convention and are in the process of testing a control group of sighted people. This effort would have been impossible without a source of support for interdisciplinary projects, Bedny says. "You just can't do this kind of research without an interdisciplinary team because you need so many different kinds of expertise," from linguistics to neuroimaging to TMS. "We really needed the whole team to make it happen."
In another example of institute-funded brain research, neuroscientist David Foster, of the School of Medicine, is taking on perhaps the most basic of all aspects of learning: memory formation. Specifically, Foster is interested in how certain kinds of memories are formed in a brain region called the hippocampus. He has studied this process in detail in rat brains, using dozens of implanted electrodes to precisely record electrical signals as the rats' neurons fire in sequences that represent stored memories. Foster would like to carry out similar studies in humans, but he cannot just go sticking electrodes deep into people's brains. So he first needs to develop less-invasive procedures.
Foster and William Anderson, an associate professor of neurosurgery in the School of Medicine, are now developing such techniques, piggybacking on research that Anderson's group does on epilepsy patients wherein they collect and analyze electrical data gathered from the surface of the brain. By piloting their study on a small sample of patients, the researchers hope to strengthen their position for applying for a larger grant, possibly from the NIH.
Bedny and Foster, both assistant professors, say that institute funding has allowed them to take on projects that might have otherwise been too risky and uncertain for an untenured faculty member. "I probably would not do too much looking outside of my own area to collaborate if I wasn't pushed and incentivized to do so by this kind of mechanism," Foster says. "This allows me, and pays me, to invest in thinking outside of my own small area."
The research grant program is the Science of Learning Institute's first major initiative, and many of the projects from the initial funding round are close to reporting results. The institute plans to continue awarding grants for at least three more years, and possibly more, depending on funding. To assess the program's success, Landau and Fisher are tracking metrics such as publications that awardees produce and external awards that leverage institute-funded work.
The institute also just launched its second big initiative: the Distinguished Science of Learning Fellowship Program. This program will award around five postdoctoral and predoctoral fellowships annually to students wanting to pursue interdisciplinary research in learning. Each fellow will have two advisers from different disciplines.
The fellows also will play a key role in the third prong of the institute's mission: translating and disseminating results beyond academia. Traditionally, much of the learning that occurs in the nation's formal classrooms and more informal settings is not as informed by research as it could be, says Fisher. To help change that, the Science of Learning Institute recently launched partnerships with the Port Discovery Children's Museum in Baltimore and the Children's Museum of Manhattan in New York to develop exhibits that are based on the research into the science of learning. The institute also plans to hire a dissemination expert to help translate research results into classrooms and other learning settings.
The Science of Learning Institute's stated mission is "to understand and optimize the most essential part of our human capital: the ability to learn." The mission makes the institute a crucial catalyst at a university—a place dedicated to learning—where all the pieces are already in place to make major progress on one of the most important scientific questions of our time, says Landau. "One of the goals of the Science of Learning Institute," she says, "is really to sew together the parts of the university that haven't yet interacted—to make it, in President Daniels' words, one university."