MIT Latest News
MIT has issued a set of reports today outlining its progress developing the essential elements of the new MIT Stephen A. Schwarzman College of Computing.
The reports summarize the efforts of five working groups which, over the last few months, have been studying ideas and options for the college, including its structure, curriculum, faculty appointment and hiring practices, social responsibilities, and computing infrastructure. The working groups have been informed by a series of community forums; further feedback from the MIT community is now sought in response to the reports.
The Institute announced in October 2018 the creation of the MIT Schwarzman College of Computing, which represents the biggest institutional change to MIT since 1950. MIT is largely structured around five broad-reaching schools that are the Institute’s main sites for undergraduate and graduate education, and research.
In response to the pervasiveness of computing in society and academic inquiry, the MIT Schwarzman College of Computing will serve as a campus-wide “bridge” across disciplines. It will advance research in computing and computer science — especially in artificial intelligence — and enhance our understanding of the social and ethical implications of technology.
Working on solutions
The working groups consist of over 100 MIT faculty, students, and staff, and have been in operation since February, with the help of community input and a campus-wide Idea Bank. The groups each submitted separate reports last week.
The working group co-chairs are also part of a steering committee which is helping guide the formation of the new college and has convened frequently in recent months to examine overlapping areas of interest among the groups. Steering committee members also include MIT Provost Martin A. Schmidt, Dean of Engineering Anantha Chandrakasan, and Faculty Chair Susan Silbey.
“I wish to express my deep appreciation to the Steering Committee and to all of the members of the working groups for their dedicated work during the last several months, especially knowing that they had a great deal of territory to cover during a relatively short span of time,” said Schmidt in an email sent to the MIT community today. “We are extremely grateful for their efforts.”
Each working group evaluated multiple, often overlapping ideas about the Schwarzman College of Computing. These working group reports do not represent a series of final decisions about the college; rather, they detail important organizational options, often weighing pros and cons of particular ideas.
The Working Group on Organizational Structure was chaired by Asu Ozdaglar, head of the Department of Electrical Engineering and Computer Science (EECS) and the School of Engineering Distinguished Professor of Engineering, and Nelson Repenning, associate dean of leadership and special projects and the Sloan School of Management Distinguished Professor of System Dynamics and Organization Studies.
The group evaluated the best organizational structure for the MIT Schwarzman College of Computing in light of the existing strengths of computing research in EECS and the overall needs of MIT’s five schools: the School of Engineering; the School of Science; the School of Humanities, Arts, and Social Sciences; the School of Architecture and Planning; and the Sloan School of Management.
The working group discussed a structure in which all five schools work to create interdisciplinary core course offerings in the new college. Another key issue the group has been examining is the relationship between the college and EECS. Additionally, the group outlined several ways that faculty can be affiliated with the college while continuing as members of their own departments and programs.
The Faculty Appointments Working Group was co-chaired by Eran Ben-Joseph, head of the Department of Urban Studies and Planning, and William Freeman, the Thomas and Gerd Perkins Professor of Electrical Engineering.
The group examined options concerning four related topics: types of faculty appointments, hiring models, faculty rights and responsibilities, and faculty mentoring handbooks. Many faculty hires could be joint appointments, the group proposed, with teaching and research in both the new college and existing departments; the college’s hiring process could also allow for a significant portion of new faculty to have this kind of multidisciplinary status.
If this approach is followed, the working group suggested, joint-faculty roles, rights and obligations need to be well-defined — including research expectations and teaching commitments — and guidelines for faculty mentoring should be established in advance.
The Working Group on Curriculum and Degrees was co-chaired by Srini Devadas, the Edwin Sibley Webster Professor of Electrical Engineering and Computer Science, and Troy Van Voorhis, the Haslam and Dewey Professor of Chemistry.
Proposals from this group include ways to encourage more undergraduates to complete the flexible computer science minor or to pursue “threads” — sets of coursework similar to minors — enhancing computing studies within their own majors. MIT might continue to expand joint degrees or even more-encompassing double majors, and might consider establishing a General Institute Requirement in computing. The group also examined graduate education and developed ideas about graduate degrees and certificates in computation, as well as the expansion of joint graduate degrees that include computing. The group also outlined a variety of ways new curriculum development may occur.
The Working Group on the Social Implications and Responsibilities of Computing was co-chaired by Melissa Nobles, the Kenan Sahin Dean of Humanities, Arts, and Social Sciences and a professor of political science, and Julie Shah, an associate professor in the Department of Aeronautics and Astronautics and head of the Interactive Robotics Group in CSAIL.
Broadly, the working group examined how best to incorporate social and ethical considerations into the college’s fabric — including education, research, and external engagement. On the education front, the group examined how that stand-alone classes about ethics and social responsibility could be woven into the college curriculum. They also evaluated how smaller educational units about social issues could be incorporated within other classes. The group also proposed new ideas about including an ethics dimension in research and extracurricular learning — such as leveraging MIT’s UROP program or mentored projects to provide a strong grounding in ethics-focused work.
The Working Group on College Infrastructure was co-chaired by Benoit Forget, an associate professor in the Department of Nuclear Science and Engineering, and Nicholas Roy, a professor in the Department of Aeronautics and Astronautics and a member of CSAIL.
This working group took particularly in-depth look at MIT’s future needs in the area of computing infrastructure. The group suggested that MIT’s future computing infrastructure is unlikely to be optimized around a single model of computing access, given the diversity of research projects and needs on campus. In general, the group suggested that support for a renewed computing infrastructure and improved data management should be a high priority for the college, and might include expanded student training and increased professional staffing in computing.
The way forward
Members of the MIT community are encouraged to examine the latest reports and offer input about the MIT Schwarzman College of Computing.
“I invite you to review these preliminary reports and provide us with your feedback, Schmidt said in his letter to the community, adding: “I look forward to further opportunities for community involvement in the early phases and continuing development of our new college.”
He noted that community input will be collected until June 28, after which the final reports will be posted.
The official launch of the MIT Schwarzman College of Computing will occur this fall, with the full development of the college occurring over a period of several years. MIT aims to add 50 full-time faculty to the college and jointly with departments across MIT over a five-year period. The Institute has also identified the location for a new building for the college, on the site of 44 Vassar Street, between Massachusetts Avenue and Main Street, and aims to open the new facility by late 2022.
In February, MIT announced the appointment of Dan Huttenlocher SM ’84 PhD ’88 as the first dean of the college. Huttenlocher will begin the new post this summer.
The MIT Schwarzman College of Computing is being supported by a $1 billion commitment for new research and education in computing, the biggest investment of its kind by a U.S. academic institution. The core support for the new college comes from a $350 million foundational gift from Stephen A. Schwarzman, the chairman, CEO, and co-founder of Blackstone, the global asset management and financial services firm.
The city of Amsterdam envisions a future where fleets of autonomous boats cruise its many canals to transport goods and people, collect trash, or self-assemble into floating stages and bridges. To further that vision, MIT researchers have given new capabilities to their fleet of robotic boats — which are being developed as part of an ongoing project — that lets them target and clasp onto each other, and keep trying if they fail.
About a quarter of Amsterdam’s surface area is water, with 165 canals winding alongside busy city streets. Several years ago, MIT and the Amsterdam Institute for Advanced Metropolitan Solutions (AMS Institute) teamed up on the “Roboat” project. The idea is to build a fleet of autonomous robotic boats — rectangular hulls equipped with sensors, thrusters, microcontrollers, GPS modules, cameras, and other hardware — that provides intelligent mobility on water to relieve congestion in the city’s busy streets.
One of project’s objectives is to create roboat units that provide on-demand transporation on waterways. Another objective is using the roboat units to automatically form “pop-up” structures, such as foot bridges, performance stages, or even food markets. The structures could then automatically disassemble at set times and reform into target structures for different activities. Additionally, the roboat units could be used as agile sensors to gather data on the city’s infrastructure, and air and water quality, among other things.
In 2016, MIT researchers tested a roboat prototype that cruised around Amsterdam’s canals, moving forward, backward, and laterally along a preprogrammed path. Last year, researchers designed low-cost, 3-D-printed, one-quarter scale versions of the boats, which were more efficient and agile, and came equipped with advanced trajectory-tracking algorithms.
In a paper presented at the International Conference on Robotics and Automation, the researchers describe roboat units that can now identify and connect to docking stations. Control algorithms guide the roboats to the target, where they automatically connect to a customized latching mechanism with millimeter precision. Moreover, the roboat notices if it has missed the connection, backs up, and tries again.
The researchers tested the latching technique in a swimming pool at MIT and in the Charles River, where waters are rougher. In both instances, the roboat units were usually able to successfully connect in about 10 seconds, starting from around 1 meter away, or they succeeded after a few failed attempts. In Amsterdam, the system could be especially useful for overnight garbage collection. Roboat units could sail around a canal, locate and latch onto platforms holding trash containers, and haul them back to collection facilities.
“In Amsterdam, canals were once used for transportation and other things the roads are now used for. Roads near canals are now very congested — and have noise and pollution — so the city wants to add more functionality back to the canals,” says first author Luis Mateos, a graduate student in the Department of Urban Studies and Planning (DUSP) and a researcher in the MIT Senseable City Lab. “Self-driving technologies can save time, costs and energy, and improve the city moving forward.”
“The aim is to use roboat units to bring new capabilities to life on the water,” adds co-author Daniela Rus, director of the Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Andrew and Erna Viterbi Professor of Electrical Engineering and Computer Science. “The new latching mechanism is very important for creating pop-up structures. Roboat does not need latching for autonomous transporation on water, but you need the latching to create any structure, whether it’s mobile or fixed.”
Joining Mateos on the paper are: Wei Wang, a joint postdoc in CSAIL and the Senseable City Lab; Banti Gheneti, a graduate student in the Department of Electrical Engineering and Computer Science; Fabio Duarte, a DUSP and Senseable City Lab research scientist; and Carlo Ratti, director of the Senseable City Lab and a principal investigator and professor of the practice in DUSP.
Making the connection
Each roboat is equipped with latching mechanisms, including ball and socket components, on its front, back, and sides. The ball component resembles a badminton shuttlecock — a cone-shaped, rubber body with a metal ball at the end. The socket component is a wide funnel that guides the ball component into a receptor. Inside the funnel, a laser beam acts like a security system that detects when the ball crosses into the receptor. That activates a mechanism with three arms that closes around and captures the ball, while also sending a feedback signal to both roboats that the connection is complete.
On the software side, the roboats run on custom computer vision and control techniques. Each roboat has a LIDAR system and camera, so they can autonomously move from point to point around the canals. Each docking station — typically an unmoving roboat — has a sheet of paper imprinted with an augmented reality tag, called an AprilTag, which resembles a simplified QR code. Commonly used for robotic applications, AprilTags enable robots to detect and compute their precise 3-D position and orientation relative to the tag.
Both the AprilTags and cameras are located in the same locations in center of the roboats. When a traveling roboat is roughly one or two meters away from the stationary AprilTag, the roboat calculates its position and orientation to the tag. Typically, this would generate a 3-D map for boat motion, including roll, pitch, and yaw (left and right). But an algorithm strips away everything except yaw. This produces an easy-to-compute 2-D plane that measures the roboat camera’s distance away and distance left and right of the tag. Using that information, the roboat steers itself toward the tag. By keeping the camera and tag perfectly aligned, the roboat is able to precisely connect.
The funnel compensates for any misalignment in the roboat’s pitch (rocking up and down) and heave (vertical up and down), as canal waves are relatively small. If, however, the roboat goes beyond its calculated distance, and doesn’t receive a feedback signal from the laser beam, it knows it has missed. “In challenging waters, sometimes roboat units at the current one-quarter scale, are not strong enough to overcome wind gusts or heavy water currents,” Mateos says. “A logic component on the roboat says, ‘You missed, so back up, recalculate your position, and try again.’”
The researchers are now designing roboat units roughly four times the size of the current iterations, so they’ll be more stable on water. Mateos is also working on an update to the funnel that includes tentacle-like rubber grippers that tighten around the pin — like a squid grasping its prey. That could help give the roboat units more control when, say, they’re towing platforms or other roboats through narrow canals.
In the works is also a system that displays the AprilTags on an LCD monitor that changes codes to signal multiple roboat units to assemble in a given order. At first, all roboat units will be given a code to stay exactly a meter apart. Then, the code changes to direct the first roboat to latch. After, the screen switches codes to order the next roboat to latch, and so on. “It’s like the telephone game. The changing code passes a message to one roboat at a time, and that message tells them what to do,” Mateos says.
Darwin Caldwell, the research director of Advanced Robotics at the Italian Institute of Technology, envisions even more possible applications for the autonomous latching capability. “I can certainly see this type of autonomous docking being of use in many areas of robotic ‘refuelling’ and docking … beyond aquatic/naval systems,” he says, “including inflight refuelling, space docking, cargo container handling, [and] robot in-house recharging.”
The research was funded by the AMS Institute and the City of Amsterdam.
Almost every day, news headlines announce another security breach and the theft of credit card numbers and other personal information. While having one’s credit card stolen can be annoying and unsettling, a far more significant, yet less recognized, concern is the security of physical infrastructure, including energy systems.
“With a credit card theft, you might have to pay $50 and get a new credit card,” says Stuart Madnick, the John Norris Maguire Professor of Information Technologies at the Sloan School of Management, a professor of engineering systems at the School of Engineering, and founding director of the Cybersecurity at MIT Sloan consortium. “But with infrastructure attacks, real physical damage can occur, and recovery can take weeks or months.”
A few examples demonstrate the threat. In 2008, an alleged cyberattack blew up an oil pipeline in Turkey, shutting it down for three weeks; in 2009, the malicious Stuxnet computer worm destroyed hundreds of Iranian centrifuges, disrupting that country’s nuclear fuel enrichment program; and in 2015, an attack brought down a section of the Ukrainian power grid — for just six hours, but substations on the grid had to be operated manually for months.
According to Madnick, for adversaries to mount a successful attack, they must have the capability, the opportunity, and the motivation. In recent incidents, all three factors have aligned, and attackers have crippled major physical systems.
“The good news is that, at least in the United States, we haven’t really experienced that yet,” says Madnick. But he believes that “it’s only motivation that’s lacking.” Given sufficient motivation, attackers anywhere in the world could, for example, bring down some or all of the nation’s interconnected power grid or stop the flow of natural gas through the country’s 2.4 million miles of pipeline. And while emergency facilities and fuel supplies may keep things running for a few days, it’s likely to take far longer than that to repair systems that attackers have damaged or blown up.
“Those are massive impacts that would affect our day-to-day life,” says Madnick. “And it’s not on most people’s radar. But just hoping that it won’t happen is not exactly a safe way to go about life.” He firmly believes that “the worst is yet to come.”
The challenge for industry
Ensuring the cybersecurity of energy systems is a growing challenge. Why? Today’s industrial facilities rely extensively on software for plant control, rather than on traditional electro-mechanical devices. In some cases, even functions critical for ensuring safety are almost entirely implemented in software. In a typical industrial facility, dozens of programmable computing systems distributed throughout the plant provide local control of processes — for example, maintaining the water-level in a boiler at a certain setpoint. Those devices all interact with a higher-level “supervisory” system that enables operators to control the local systems and overall plant operation, either on-site or remotely. In most facilities, these programmable computing systems do not require any authentication for settings to be altered. Given this setup, a cyberattacker who gains access to the software in either the local or the supervisory system can cause damage or disruption of service.
The traditional approach used to protect critical control systems is to “air-gap” them — that is, separate them from the public internet so that intruders can’t reach them. But in today’s world of high connectivity, an air-gap no longer guarantees security. For example, companies often hire independent contractors or vendors to maintain and monitor specialized equipment in their facilities. To perform those tasks, the contractor or vendor needs access to real-time operational data — information that’s generally transmitted over the internet. In addition, legitimate business needs, such as transferring files and updating software, require the use of USB sticks, which can inadvertently jeopardize the integrity of the air-gap, leaving a plant vulnerable to cyberattack.
Looking for vulnerabilities
Companies actively work to tighten up their security — but typically only after some incident has occurred. “So we tend to be looking through the rear-view mirror,” says Madnick. He stresses the need to identify and mitigate the vulnerabilities of a system before a problem arises.
The traditional method of identifying cyber-vulnerabilities is to create an inventory of all the components, examine each one to identify any vulnerabilities, mitigate those vulnerabilities, and then aggregate the results to secure the overall system. But that approach relies on two key simplifying assumptions, says Shaharyar Khan, a fellow of the MIT System Design and Management program. It assumes that events always run in a single, linear direction, so one event causes another event, which causes another event, and so on, without feedback loops or interactions to complicate the sequence. And it assumes that understanding the behavior of each component in isolation is sufficient to predict the behavior of the overall system.
But those assumptions don’t hold for complex systems — and modern control systems in energy facilities are extremely complex, software-intensive, and made up of highly coupled components that interact in many ways. As a result, says Khan, “the overall system exhibits behaviors that the individual components do not” — a property known in systems theory as emergence. “We consider safety and security to be emergent properties of systems,” says Khan. The challenge is therefore to control the emergent behavior of the system by defining new constraints, a task that requires understanding how all the interacting factors at work — from people to equipment to external regulations and more — ultimately impact system safety.
To develop an analytical tool up to that challenge, Madnick, Khan, and James L. Kirtley Jr., a professor of electrical engineering, turned first to a methodology called System Theoretic Accident Model and Process, which was developed more than 15 years ago by MIT Professor Nancy Leveson of aeronautics and astronautics. With that work as a foundation, they developed “Cybersafety,” an analytical method specifically tailored for cybersecurity analysis of complex industrial control systems.
To apply the Cybersafety procedure to a facility, an analyst begins by answering the following questions:
• What is the main purpose of the system being analyzed; that is, what do you need to protect? Answering that question may sound straightforward, but Madnick notes, “Surprisingly, when we ask companies what their ‘crown jewels’ are, they often have trouble identifying them.”
• Given that main purpose, what’s the worst that could happen to the system? Defining the main purpose and the worst possible losses is key to understanding the goal of the analysis and the best allocation of resources for mitigation.
• What are key hazards that could lead to that loss? As a simple example, having wet stairs in a facility is a hazard; having someone fall down the stairs and break an ankle is a loss.
• Who or what controls that hazard? In the above example, the first step is to determine who or what controls the state of the stairs. The next step is to ask, Who or what controls that controller? And then, Who or what controls that controller? Answering that question recursively and mapping the feedback loops among the various controllers yields a hierarchical control structure responsible for maintaining the state of the stairs in an acceptable condition.
Given the full control structure, the next step is to ask: What control actions might be taken by a controller that would be unsafe given the state of the system? For example, if an attacker corrupts feedback from a key sensor, a controller will not know the actual state of the system and therefore may take an incorrect action, or may take the correct actions but at the wrong time or in the wrong order — any of which would lead to damage.
Based on the now-deeper understanding of the system, the analyst next hypothesizes a series of loss scenarios stemming from unsafe control actions and examines how the various controllers might interact to issue an unsafe command. “At each level of the analysis, we try to identify constraints on the process being controlled that, if violated, would result in the system moving into an unsafe state,” says Khan. For example, one constraint could dictate that the steam pressure inside a boiler must not exceed a certain upper bound to prevent the boiler from bursting due to over-pressure.
“By continually refining those constraints as we progress through the analysis, we are able to define new requirements that will ensure the safety and security of the overall system,” he says. “Then we can identify practical steps for enforcing adherence to those constraints through system design, processes and procedures, or social controls such as company culture, regulatory requirements, or insurance incentives.”
To demonstrate the capabilities of Cybersafety analysis, Khan selected a 20-megawatt, gas turbine power plant — a small facility that has all the elements of a full-scale power plant on the grid. In one analysis, he examined the control system for the gas turbine, focusing in particular on how the software controlling the fuel-control valve could be altered to cause system-level losses.
Performing the Cybersafety analysis yielded several turbine-related loss scenarios involving fires or explosions, catastrophic equipment failure, and ultimately the inability to generate power.
For example, in one scenario, the attacker disables the turbine’s digital protection system and alters the logic in the software that controls the fuel-control valve to keep the valve open when it should be closed, stopping fuel from flowing into the turbine. If the turbine is then suddenly disconnected from the grid, it will begin to spin faster than its design limit and will break apart, damaging nearby equipment and harming workers in the area.
The Cybersafety analysis uncovered the source of that vulnerability: An updated version of the control system had eliminated a backup mechanical bolt assembly that ensured turbine “over-speed” protection. Instead, over-speed protection was implemented entirely in software.
That change made sense from a business perspective. A mechanical device requires regular maintenance and testing, and those tests subject the turbine to such extreme stresses that it sometimes fails. However, given the importance of cybersecurity, it might be wise to bring back the mechanical bolt as a standalone safety device — or at least to consider standalone electronic over-speed protection schemes as a final line of defense.
Another case study focused on systems used to deliver chilled water and air conditioning to the buildings being served. Once again, the Cybersafety analysis revealed multiple loss scenarios; in this case, most had one cause in common: the use of variable frequency drives (VFDs) to adjust the speed of motors that drive water pumps and compressors.
Like all motors, the motor driving the chiller’s compressor has certain critical speeds at which mechanical resonance occurs, causing excessive vibration. VFDs are typically programmed to skip over those critical speeds during motor startup. But some VFDs are programmable over the network. Thus, an attacker can query a VFD for the critical speed of the attached motor and then command it to drive the motor at that dangerous speed, permanently damaging it.
“This is a simple kind of an attack; it doesn’t require a lot of sophistication,” says Khan. “But it could be launched and could cause catastrophic damage.” He cites earlier work performed by Matthew Angle ’07, MEng ’11, PhD ’16, in collaboration with Madnick and Kirtley. As part of a 2017 study of cyberattacks on industrial control systems, Angle built a lab-scale motor test kit equipped with a complete VFD with computer code familiar to the researchers. By simply altering a few key lines of code, they caused capacitors in the VFD to explode, sending smoke billowing out into the courtyard behind their MIT lab. In an industrial setting with full-sized VFDs, a similar cyberattack could cause significant structural damage and potentially harm personnel.
Given such possibilities, the research team recommends that companies carefully consider the “functionality” of the equipment in their system. Many times, plant personnel are not even aware of the capabilities that their equipment offers. For example, they may not realize that a VFD driving a motor in their plant can be made to operate in reverse direction by a small change in the computer code controlling it — a clear cyber-vulnerability. Removing that vulnerability would require using a VFD with less functionality. “Good engineering to remove such vulnerabilities can sometimes be mistakenly characterized as a move backwards, but it may be necessary to improve a plant’s security posture,” says Khan. A full Cybersafety analysis of a system will not only highlight such issues, but also guide the strategic placement of analog sensors and other redundant feedback loops that will increase the resiliency of system operation.
Addressing the challenge
Throughout their cybersecurity research, Khan, Madnick, and their colleagues have found that vulnerabilities can often be traced to human behavior, as well as management decisions. In one case, a company had included the default passcode for its equipment in the operator’s manual, publicly available on the internet. Other cases involved operators connecting USB drives and personal laptops directly into the plant network, thereby breaching the air-gap and even introducing malware into the plant control system.
In one case, an overnight worker downloaded movies onto a plant computer using a USB stick. But often such actions were taken as part of desperate attempts to get a currently shut-down plant back up and running. “In the grand scheme of priorities, I understand that focusing on getting the plant running again is part of the culture,” says Madnick. “Unfortunately, the things people do in order to keep their plant running sometimes puts the plant at an even greater risk.”
Enabling a new culture and mindset requires a serious commitment to cybersecurity up the management chain. Mitigation strategies are likely to call for reengineering the control system, buying new equipment, or making changes in processes and procedures that might incur extra costs. Given what’s at stake, management must not only approve such investments, but also instill a sense of urgency in their organizations to identify vulnerabilities and eliminate or mitigate them.
Based on their studies, the researchers conclude that it’s impossible to guarantee that an industrial control system will never have its network defenses breached. “Therefore, the system must be designed so that it’s resilient against the effects of an attack,” says Khan. “Cybersafety analysis is a powerful method because it generates a whole set of requirements — not just technical but also organizational, logistical, and procedural — that can improve the resilience of any complex energy system against a cyberattack.”
This research was supported by the U.S. Department of Energy, the MIT Energy Initiative Seed Fund Program, and members of the Cybersecurity at MIT Sloan consortium. More information and the latest publications can be found at cams.mit.edu.
This article appears in the Spring 2019 issue of Energy Futures, the magazine of the MIT Energy Initiative.
The past four decades have been transformative for manufacturing. An explosive growth of new technologies has revolutionized how products are made and distributed. In the 1980s, the steep rise in Japanese manufacturing reshaped the global market. Advances in the fields of automation, robotics, and factory systems have drastically altered the landscape of the traditional factory floor. David Hardt, the Ralph E. and Eloise F. Cross Professor in Manufacturing, has had a front-row seat to these radical changes.
Hardt SM ’75, PhD ’78 joined the faculty in MIT’s Department of Mechanical Engineering (MechE) in 1979 and later served as director of the MIT Laboratory for Manufacturing for nine years. A leading expert in manufacturing process control, Hardt pioneered new equipment and control techniques in fields such as gas metal arc welding, metal forming in the aerospace industry, and micro-fluidic device manufacture.
Hardt was also involved with the initiation and management of the “Leaders for Manufacturing” (now LGO) program, a collaboration between the MIT Sloan School of Management and the School of Engineering, serving as engineering co-director for four years. From this and his MechE work, he noticed that MIT’s degree programs weren’t adequately preparing engineering students for careers in manufacturing. In 2010, he helped develop MIT’s master of engineering in advanced manufacturing and design (MEngM), a yearlong program that prepares graduate students to be engineering leaders in manufacturing.
Last year, Hardt and colleagues like Sanja Sarma, vice president for open learning, took lessons from the MEngM degree and launched the MITx Micromasters Program in Principles of Manufacturing, on online program about the fundamentals of manufacturing as developed in the MEngM.
Q: How did you decide to spend your career focusing on manufacturing?
A: Well, when I got to MIT I was enamored with biomedical engineering. I studied muscle-force control during walking for my PhD. By the time I graduated, the market was oversaturated and no one was interested in hiring a biomedical engineer. So I took a postdoctoral role in manufacturing at MIT. I had studied control and dynamics in graduate school and started thinking of ways I could apply that to manufacturing. That’s when I took the theme of process control and ran with it. In the parlance of controls, I was expanding the control to include the whole manufacturing process — not just the machine itself.
Q: In the 40 years since you joined the faculty, what has been the biggest change you’ve seen in manufacturing?
A: The level of sophistication has been one of the biggest changes. Manufacturing has become such a highly refined activity globally. Look at any modern manufacturing operation and it has to be one of the most complex technical systems there are on earth.
It used to be that with enough labor, some skill, space, and time, you could make anything and make a profit. But the standards manufacturers are now held to are extremely high. You can’t make something with poor quality and high cost and get away with it anymore. Consumers’ expectations have really upped the ante.
Q: Rethinking how manufacturing is taught has been a theme throughout your career. How did the MEngM program initially come to fruition?
A: I started collaborating more with colleagues from Sloan School of Management, as well as managers and operating engineers in industry. It gave me more of a ground truth in what was important in manufacturing. That opened my eyes and in some of the classes I was teaching, I shifted from a purely mechanical engineering approach to a broader, more pragmatic approach that took into account what was really happening in industry.
When the [Singapore-MIT Alliance for Research and Technology] began in 1998, we knew we wanted to collaborate with researchers in Singapore on manufacturing. We developed a novel professional manufacturing degree program in Singapore. For five years, we ran it from a distance. It was a roaring success, so we realized that there was an opportunity to start a similar program right here at MIT, and launched the MEngM program. For our students, it’s like a capstone degree. Undergraduate manufacturing classes just scratch the surface — the MEngM really educates students in the theory and practice of manufacturing.
Q: How did you use the lessons you’ve learned from the MEngM program to shape the MITx Micromasters Program in Principles of Manufacturing?
A: There are four core classes in the MEngM program that we started calling the "principles of manufacturing." We realized that teaching those classes as a unit would provide great utility on their own. Someone working in industry who has a mechanical engineering background could take those classes and it would greatly enhance their ability to work in manufacturing and design. So, along with my colleagues Jung-Hoon Chun, Stephen Graves, Duane Boning, Stan Gershwin, Jose Pacheco, and John Liu, I worked with Professor Sanjay Sarma and the MIT edX team to put together eight online courses on manufacturing process control, manufacturing systems, management in engineering, and supply chains for manufacturing. The courses are taught by a seasoned team of faculty from MIT MechE, MIT Leaders for Global Operations Program, and Sloan School of Management.
Q: What are you hoping students will take away from the Micromasters Program?
A: Everybody knows that the biggest hurdle in manufacturing is the conversion from a groundbreaking idea to actual production. We hope that the program can help professionals across industry surmount that hurdle. Our first year of the program just launched in March 2018, and we have had students from all across the world at varying levels in their career. Our first Micromasters credential should be awarded this fall, and we hope to admit some of them to the MEngM. I’m looking forward to hearing more from them about how they plan to implement the skills they learned through the program throughout their careers.
The School of Engineering has announced that 17 members of its faculty have been granted tenure by MIT.
“The tenured faculty in this year’s cohort are a true inspiration,” said Anantha Chandrakasan, dean of the School of Engineering. “They have shown exceptional dedication to research and teaching, and their innovative work has greatly advanced their fields.”
This year’s newly tenured associate professors are:
Antoine Allanore, in the Department of Materials Science and Engineering, develops more sustainable technologies and strategies for mining, metal extraction, and manufacturing, including novel methods of fertilizer production.
Saurabh Amin, in the Department of Civil and Environmental Engineering, focuses on the design and implementation of network inspection and control algorithms for improving the resilience of large-scale critical infrastructures, such as transportation systems and water and energy distribution networks, against cyber-physical security attacks and natural events.
Emilio Baglietto, in the Department of Nuclear Science and Engineering, uses computational modeling to characterize and predict the underlying heat-transfer processes in nuclear reactors, including turbulence modeling, unsteady flow phenomena, multiphase flow, and boiling.
Paul Blainey, the Karl Van Tassel (1925) Career Development Professor in the Department of Biological Engineering, integrates microfluidic, optical, and molecular tools for application in biology and medicine across a range of scales.
Kerri Cahoy, the Rockwell International Career Development Professor in the Department of Aeronautics and Astronautics, develops nanosatellites that demonstrate weather sensing using microwave radiometers and GPS radio occultation receivers, high data-rate laser communications with precision time transfer, and active optical imaging systems using MEMS deformable mirrors for exoplanet exploration applications.
Juejun Hu, in the Department of Materials Science and Engineering, focuses on novel materials and devices to exploit interactions of light with matter, with applications in on-chip sensing and spectroscopy, flexible and polymer photonics, and optics for solar energy.
Sertac Karaman, the Class of 1948 Career Development Professor in the Department of Aeronautics and Astronautics, studies robotics, control theory, and the application of probability theory, stochastic processes, and optimization for cyber-physical systems such as driverless cars and drones.
R. Scott Kemp, the Class of 1943 Career Development Professor in the Department of Nuclear Science and Engineering, combines physics, politics, and history to identify options for addressing nuclear weapons and energy. He investigates technical threats to nuclear-deterrence stability and the information theory of treaty verification; he is also developing technical tools for reconstructing the histories of secret nuclear-weapon programs.
Aleksander Mądry, in the Department of Electrical Engineering and Computer Science, investigates topics ranging from developing new algorithms using continuous optimization, to combining theoretical and empirical insights, to building a more principled and thorough understanding of key machine learning tools. A major theme of his research is rethinking machine learning from the perspective of security and robustness.
Frances Ross, the Ellen Swallow Richards Professor in the Department of Materials Science and Engineering, performs research on nanostructures using transmission electron microscopes that allow researchers to see, in real-time, how structures form and develop in response to changes in temperature, environment, and other variables. Understanding crystal growth at the nanoscale is helpful in creating precisely controlled materials for applications in microelectronics and energy conversion and storage.
Daniel Sanchez, in the Department of Electrical Engineering and Computer Science, works on computer architecture and computer systems, with an emphasis on large-scale multi-core processors, scalable and efficient memory hierarchies, architectures with quality-of-service guarantees, and scalable runtimes and schedulers.
Themistoklis Sapsis, the Doherty Career Development Professor in the Department of Mechanical Engineering, develops analytical, computational, and data-driven methods for the probabilistic prediction and quantification of extreme events in high-dimensional nonlinear systems such as turbulent fluid flows and nonlinear mechanical systems.
Julie Shah, the Boeing Career Development Professor in the Department of Aeronautics and Astronautics, develops innovative computational models and algorithms expanding the use of human cognitive models for artificial intelligence. Her research has produced novel forms of human-machine teaming in manufacturing assembly lines, healthcare applications, transportation, and defense.
Hadley Sikes, the Esther and Harold E. Edgerton Career Development Professor in the Department of Chemical Engineering, employs biomolecular engineering and knowledge of reaction networks to detect epigenetic modifications that can guide cancer treatment, induce oxidant-specific perturbations in tumors for therapeutic benefit, and improve signaling reactions and assay formats used in medical diagnostics.
William Tisdale, the ARCO Career Development Professor in the Department of Chemical Engineering, works on energy transport in nanomaterials, nonlinear spectroscopy, and spectroscopic imaging to better understand and control the mechanisms by which excitons, free charges, heat, and reactive chemical species are converted to more useful forms of energy, and on leveraging this understanding to guide materials design and process optimization.
Virginia Vassilevska Williams, the Steven and Renee Finn Career Development Professor in the Department of Electrical Engineering and Computer Science, applies combinatorial and graph theoretic tools to develop efficient algorithms for matrix multiplication, shortest paths, and a variety of other fundamental problems. Her recent research is centered on proving tight relationships between seemingly different computational problems. She is also interested in computational social choice issues, such as making elections computationally resistant to manipulation.
Amos Winter, the Tata Career Development Professor in the Department of Mechanical Engineering, focuses on connections between mechanical design theory and user-centered product design to create simple, elegant technological solutions for applications in medical devices, water purification, agriculture, automotive, and other technologies used in highly constrained environments.
Objects made with 3-D printing can be lighter, stronger, and more complex than those produced through traditional manufacturing methods. But several technical challenges must be overcome before 3-D printing transforms the production of most devices.
Commercially available printers generally offer only high speed, high precision, or high-quality materials. Rarely do they offer all three, limiting their usefulness as a manufacturing tool. Today, 3-D printing is used mainly for prototyping and low-volume production of specialized parts.
Now Inkbit, a startup out of MIT, is working to bring all of the benefits of 3-D printing to a slew of products that have never been printed before — and it’s aiming to do so at volumes that would radically disrupt production processes in a variety of industries.
The company is accomplishing this by pairing its multimaterial inkjet 3-D printer with machine-vision and machine-learning systems. The vision system comprehensively scans each layer of the object as it’s being printed to correct errors in real-time, while the machine-learning system uses that information to predict the warping behavior of materials and make more accurate final products.
“The company was born out of the idea of endowing a 3-D printer with eyes and brains,” says Inkbit co-founder and CEO Davide Marini PhD ’03.
That idea unlocks a range of applications for Inkbit’s machine. The company says it can print more flexible materials much more accurately than other printers. If an object, including a computer chip or other electronic component, is placed on the print area, the machine can precisely print materials around it. And when an object is complete, the machine keeps a digital replica that can be used for quality assurance.
Inkbit is still an early-stage company. It currently has one operational production-grade printer. But it will begin selling printed products later this year, starting with a pilot with Johnson and Johnson, before selling its printers next year. If Inkbit can leverage current interest from companies that sell medical devices, consumer products, and automotive components, its machines will be playing a leading production role in a host of multi-billion-dollar markets in the next few years, from dental aligners to industrial tooling and sleep apnea masks.
“Everyone knows the advantages of 3-D printing are enormous,” Marini says. “But most people are experiencing problems adopting it. The technology just isn’t there yet. Our machine is the first one that can learn the properties of a material and predict its behavior. I believe it will be transformative, because it will enable anyone to go from an idea to a usable product extremely quickly. It opens up business opportunities for everyone.”
A printer with potential
Some of the hardest materials to print today are also the most commonly used in current manufacturing processes. That includes rubber-like materials such as silicone, and high-temperature materials such as epoxy, which are often used for insulating electronics and in a variety of consumer, health, and industrial products.
These materials are usually difficult to print, leading to uneven distribution and print process failures like clogging. They also tend to shrink or round at the edges over time. Inkbit co-founders Wojciech Matusik, an associate professor of electrical engineering and computer science, Javier Ramos BS ’12 SM ’14, Wenshou Wang, and Kiril Vidimče SM ’14 have been working on these problems for years in Matusik’s Computational Fabrications Group within the Computer Science and Artificial Intelligence Laboratory (CSAIL).
In 2015, the co-founders were among a group of researchers that created a relatively low-cost, precise 3-D printer that could print a record 10 materials at once by leveraging machine vision. The feat got the attention of many large companies interested in transitioning production to 3-D printing, and the following year the four engineers received support from the Deshpande Center to commercialize their idea of joining machine vision with 3-D printing.
At MIT, Matusik’s research group used a simple 3-D scanner to track its machine’s progress. For Inkbit’s first printer, the founders wanted to dramatically improve “the eyes” of their machine. They decided to use an optical coherence tomography (OCT) scanner, which uses long wavelengths of light to see through the surface of materials and scan layers of material at a resolution the fraction of the width of a human hair.
Because OCT scanners are traditionally only used by ophthalmologists to examine below the surface of patients’ eyes, the only ones available were far too slow to scan each layer of a 3-D printed part — so Inkbit’s team “bit the bullet,” as Marini describes it, and built a custom OCT scanner he says is 100 times faster than anything else on the market today.
When a layer is printed and scanned, the company’s proprietary machine-vision and machine-learning systems automatically correct any errors in real-time and proactively compensate for the warping and shrinkage behavior of a fickle material. Those processes further expand the range of materials the company is able to print with by removing the rollers and scrapers used by some other printers to ensure precision, which tend to jam when used with difficult-to-print materials.
The system is designed to allow users to prototype and manufacture new objects on the same machine. Inkbit’s current industrial printer has 16 print heads to create multimaterial parts and a print block big enough to produce hundreds of thousands of fist-sized products each year (or smaller numbers of larger products). The machine’s contactless inkjet design means increasing the size of later iterations will be as simple as expanding the print block.
“Before, people could make prototypes with multimaterial printers, but they couldn’t really manufacture final parts,” Matusik says, noting that the postprocessing of Inkbit’s parts can be fully automated. “This is something that’s not possible using any other manufacturing methods.”
Inkbit's 3-D printer can produce multimaterial objects (like the pinch valve shown above) at high volumes. Courtesy of Inkbit
The novel capabilities of Inkbit’s machine mean that some of the materials the founders want to print with are not available, so the company has created some of its own chemistries to push the performance of their products to the limit. A proprietary system for mixing two materials just before printing will be available on the printers Inkbit ships next year. The two-part chemistry mixing system will allow the company to print a broader range of engineering-grade materials.
Johnson and Johnson, a strategic partner of Inkbit, is in the process of acquiring one of the first printers. The MIT Startup Exchange Accelerator (STEX25) has also been instrumental in exposing Inkbit to leading corporations such as Amgen, Asics, BAE Systems, Bosch, Chanel, Lockheed Martin, Medtronic, Novartis, and others.
Today, the founders spend a lot of their time educating product design teams that have never been able to 3-D print their products before — let alone incorporate electronic components into 3-D-printed parts.
It may be a while before designers and inventors take full advantage of the possibilities unlocked by integrated, multimaterial 3-D printing. But for now, Inkbit is working to ensure that, when that future comes, the most imaginative people will have a machine to work with.
“Some of this is so far ahead of its time,” Matusik says. “I think it will be really fascinating to see how people are going to use it for final products.”
The Department of Civil and Environmental Engineering gathered recently to acknowledge the close of the academic year and celebrate the Class of 2019 and notable members of the CEE community. The annual event unites students, postdocs, faculty, and staff and is a great evening to reflect on the accomplishments of the year and show appreciation for the people who make CEE an outstanding department.
The graduating seniors kicked off the event by presenting the findings of their capstone projects. The CEE capstone, a component of 1.013 (Senior Civil and Environmental Engineering Design), gives seniors the opportunity to work individually or in a pair in order to conduct engineering work with a real-world impact during the final semester of their MIT undergraduate career.
The design-focused work was presented in the form of digital posters, which allowed the community to interact with each student, or pair, to learn about their projects, and for CEE faculty to evaluate and vote on the top three posters. Topics ranged from tackling climate change issues and nature-inspired materials to data analysis of transportation systems and computational toolkits for green-space design. Markus Buehler, head of CEE and McAfee Professor of Engineering, announced that first place was awarded to Apisada "Ju" Chulakadabba, while Tim Roberts earned runner-up and David Wu came in third place.
Chulakadabba’s capstone project compared global climate models to the MIT regional climate model to examine projected climate-change impacts on hydrological cycles in China. The primary areas Chulakadabba focused on were the Yangtze River Basin, where the water supply is abundant, and the Yellow River Basin, where water is scarce. Her findings from the comparison provided the future trends of the hydrological processes in China, and also evaluated the performance of the selected models in the regions. Chulakadabba’s work suggests that there is an increase in annual precipitation, runoff, and evaporation trends; nevertheless, she emphasized that regardless of the potential increase in water availability, it is still important to have the Water Transfer Project as a backup plan. Chulakadabba stressed that based on her work, the project would be justified from an environmental engineering perspective. However, it is not financially sustainable.
Shifting from environmental engineering challenges to nature-inspired materials, Tim Roberts presented his project on synthetic silk production — a promising, yet challenging, design issue. The current method occurs in live cells and can take up to five days without yielding the desired results. His project focused on designing a screening process using cell-free protein expression to assess the feasibility of producing proteins in live cell expression.
Working with systems and data, David Wu's capstone project used data analysis to evaluate the effect that Red Sox baseball games have on congestion, specifically at the Kenmore MBTA stop in Boston. After games, there is a mass exodus of people attempting to utilize MBTA transportation, whereas at the beginning of games, fans' arrival times vary. Wu analyzed the MBTA data and examined how many people use the T to leave, and how travel times are affected. Using queuing theory and the given data, he created a queuing model to simulate station operations and estimate waiting times. Wu expressed that data is often limited, and it is beneficial to learn domain-specific concepts, such as queuing theory in terms of transportation, to gain invaluable insight that statistical models cannot provide, and to design more efficient transportation strategies.
Following the capstone poster session was the presentation of the CEE awards. All recipients were nominated by both peers and advisors for being exemplary members of the community who represent the CEE mission, and significantly contribute to the department’s excellence, cutting-edge research, and education. “The awardees resemble the aspirations, values, and ideals of the MIT CEE department, recognize exceptional achievements and talents, and inspire others,” Buehler said in his opening remarks.
The first portion of awards applauded undergraduates for their dedication to the department. This year, junior Zoe Lallas received the CEE Leadership and Community Award, which recognizes an undergraduate student who makes exemplary contributions to improve the CEE community, fosters excellence and diversity, and contributes to our inclusive culture. Lallas has served as the social chair for the CEE Student Association and has been involved with the First-Year Preorientation Program, serving as a mentor one year and a student organizer the next.
Sophomore Chelsea Watanabe won the Best Undergraduate Research Award, which honors excellence in any area of research by a CEE undergraduate student, carried out in the context of either an Undergraduate Research Opportunities Program internship or through coursework, such as Traveling Research Environmental Experiences. Watanabe is known to be inspiring to work with due to her deep sense of curiosity and ambitious attitude.
Senior Christine Langston won the Leo (Class of 1924) and Mary Grossman Award for her strong interest in transportation and impressive academic record. Langston has combined data from a variety of sources such as state and local transportation agencies, Google Maps, and Trip Advisor to measure and model travel patterns within cities. Langston is recognized for her passion and drive to improve transportation systems.
Senior Tim Roberts earned the Juan Jose Hermosilla (1957) Prize for demonstrating exceptional talent and potential for future contributions at the intersection of mechanics, materials, structures, and design. Roberts was nominated for being well-rounded and for his many achievements in engineering. He is not only proficient in Spanish and Chinese, but he also performed research at several labs at MIT and completed an internship at a leading structural engineering company. Roberts’ colleagues speak highly of him, as he is known to be very humble and thoughtful, willing to go out of his way to help others.
Senior Amber VanHemel was awarded the Paul Busch (1958) Prize, given to an undergraduate student in environmental science and engineering for academic achievement and contributions to the CEE community. VanHemel is recognized by her peers and professors as an exceptionally bright, hard-working, outgoing and ambitious scholar.
Achieving the Tucker-Voss award was MEng student Andrew Novillo, who completed his thesis in experimental testing of cast-metal connections for complex loading conditions designed with topology optimization. The award was established in memory of professors Ross R. Tucker and Walter C. Voss, who were the first two department heads of the now extinct Course 17 (Building Construction). When Course 17 merged with the Department of Civil Engineering in the 1950s, the Tucker-Voss award was established. Novillo earned this award for his use of innovative 3-D printing technology in his thesis, which demonstrated the promising future he will have in the field of building.
Graduate student Hayley Gadol was awarded the Trond Kaalstad (Class of 1957) Fellowship, which recognizes an outstanding graduate student who has displayed leadership and/or contributed significantly to the well-being of the CEE community. Hayley took on the goal of improving graduate student life in the department and the Institute, serving as the head of CEE Student Graduate Committee and taking charge of organizing events for the community.
The Maseeh Annual Award for Excellence, which recognizes the most outstanding teaching assistant in the past academic year, was awarded to Hejian (Patrick) Zhu, who was an instructor for the subjects 1.361 (Advanced Soil Mechanics) and 1.364 (Advanced Geotechnical Engineering). Through his commitment as a teaching assistant, Patrick has proved to be passionate about helping others deepen their knowledge and understanding of geomechancial topics.
Receiving the Best Doctoral Thesis Award was Simone Cenci, who worked under the guidance of his advisor, Mitsui Career Development Assistant Professor in Contemporary Technology Serguei Saavedra. This award honors scholarly and academic excellence and a high level of distinction of a CEE graduate student in any area of research. Cenci produced eight impressive research papers, and has significantly contributed to the area of theoretical ecology by expanding concepts and tools that can get us closer to a better understanding and prediction of population dynamics.
The CEE Postdoctoral Scholar Mentoring, Teaching and Excellence Award recognizes mentoring, teaching, and other exceptional contributions by a postdoc, emphasizing high potential for future contributions. Ehsan Haghighat received the award for his extraordinary teaching and generous mentorship, displaying strong research in computational mechanics and more. Ehsan excelled in this teaching role by demonstrating an outstanding ability to communicate knowledge effectively to the students, as well as earning top reviews in the student evaluations.
Two members of the CEE staff received the CEE Excellence Award, which recognizes staff for excellent contributions to the community, commitment to professionalism, dedication and best practices, and for fostering a culture of diversity, inclusiveness, and innovation. The first recipient was undergraduate academic assistant Sarah Smith. Smith was acknowledged for her ability to flawlessly handle every interaction with faculty and staff with a positive, respectful attitude and a smile.
The second recipient of the CEE Excellence Award was research engineer John MacFarlane. MacFarlane is known to be a dedicated member of the department who is willing to help others, ensure safety within labs, and maintain a great attitude. Buehler noted that MacFarlane is known as a “the Life Saver” by the students, faculty, and staff.
The department also presented faculty with three awards. The Samuel M. Seegal Prize, which honors faculty members for inspiring students to pursue and achieve excellence, was awarded to William E. Leonhard Professor Harry Hemond. The CEE community noticed Hemond for being a beloved teacher and mentor who leads by example, and who inspires students long after their time at MIT. A former student wrote in the nomination: “His mentorship shaped the scientist I am today, and I continuously strive to be as knowledgeable, thorough, and creative in my work as he is,” reflecting the great impact Hemond had on his students.
Assistant Professor Lydia Bourouiba received the Ole Madsen Mentoring Award, which honors faculty members for conspicuous contributions to mentoring and educating CEE students outside the classroom, and inspiring them to pursue a career in the fields of civil and environmental engineering. Bourouiba teaches students the skills, qualities, and critical thinking required to succeed in their studies and research; more generally, she prepares them to be successful in their professional lives. One student wrote: “Her dedication and genuine care to the education, professional development, and well-being of her students and mentees are truly remarkable and extraordinary.”
Recognizing the most outstanding faculty member in the past academic year is the Maseeh Excellence in Teaching Award, which was presented to Esther and Harold E. Edgerton Career Development Assistant Professor Admir Masic. Masic stood out to his colleagues for his enthusiasm and energy for research that sparks the students’ interest in the challenge of learning. He is known by his students for his ability to make learning fun, engaging, and exciting.
“The CEE awards ceremony this year highlighted the extraordinary members of the department who contribute to our overall success, and gave the Class of 2019 an opportunity to showcase all of the hard work they have put into their capstone projects. This event exemplifies how various people in the department, from staff to the students and faculty, come together to continue fulfilling our commitment to excellence and solving important societal problems in infrastructure and environment,” Buehler says.
The following news is adapted from a press release issued by the Division for Planetary Sciences of the American Astronomical Society.
The American Astronomical Society’s Division for Planetary Sciences (DPS) has awarded the 2019 Gerard P. Kuiper Prize for outstanding contributions to the field of planetary science to MIT Professor Maria Zuber for her advancements in geophysics, planetary gravity mapping, and laser altimetry. Zuber is the E.A. Griswold Professor of Geophysics in the Department of Earth, Atmospheric and Planetary Sciences (EAPS) and vice president for research at MIT.
The Gerard P. Kuiper Prize honors scientists whose lifetime achievements have most advanced society’s understanding of the planetary system. Zuber’s numerous accomplishments include her seminal 2000 paper in Science combining Mars Global Surveyor laser altimetry data and gravity data to determine the crustal and upper mantle structure of Mars. Zuber became the first woman to lead a NASA spacecraft mission as principal investigator of the Gravity Recovery and Interior Laboratory (GRAIL) mission. GRAIL constructed a model of the moon’s gravitational field to spherical harmonic degree 1800, which exceeded the baseline requirement of the mission by an order of magnitude. Zuber has turned her attention to many different solid bodies in the solar system, focusing on structure and tectonics, including Mercury, Venus, Eros, Vesta, and Ceres. Since 1990, she has held leadership roles associated with scientific experiments or instrumentation on nine NASA missions.
Zuber has been at the helm of MIT’s research endeavors, overseeing more than a dozen interdisciplinary research laboratories and centers, ensuring intellectual integrity, and fostering research relationships. Over the years, she has advised a number of students and postdocs, and one reports that she strikes the perfect balance of being demanding, supportive, encouraging, and open-minded.
As the recipient of the prize, Zuber will be invited to present a lecture at a DPS meeting and publish a written version of it in Icarus.
On Tuesday, May 28, MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) hosted a special TEDx event featuring an all-female line-up of MIT scientists and researchers who discussed cutting-edge ideas in science and technology.
TEDxMIT speakers included roboticists, engineers, astronomers, and policy experts, including former White House chief technology officer Megan Smith ’86 SM ’88 and MIT Institute Professor Emerita Barbara Liskov, winner of the A.M. Turing Award, often considered the “Nobel Prize for computing.”
From Professor Nergis Mavalvala’s work on gravitational waves to Associate Provost Krystyn Van Vliet’s efforts to improve cell therapy, the afternoon was filled with energizing and historic success stories of women in STEM.
In an early talk, MIT Associate Professor Julie Shah touched on the much-discussed narrative of artificial intelligence and job displacement, and how that relates to her own work creating systems that she described as “being intentional about augmenting human capabilities.”She spoke about her efforts developing robots to help reduce the cognitive burden of overwhelmed workers, like the nurses on labor wards who have to make hundreds of split-second decisions for scheduling deliveries and C-sections.
“We can create a future where we don’t have robots who replace humans, but that help us accomplish what neither group can do alone,” said Shah.
CSAIL Director Daniela Rus, a professor of electrical engineering and computer science, spoke of how computer scientists can inspire the next generation of programmers by emphasizing the many possibilities that coding opens up.
“I like to say that those of us who know how to ... breathe life into things through programming have superpowers,” said Rus.
Throughout the day scientists showed off technologies that could fundamentally transform many industries, from Professor Dava Newman’s prototype Mars spacesuit to Associate Professor Vivienne Sze’s low-power processors for machine learning.
Judy Brewer, director of the World Wide Web Consortium’s Web Accessibility Initiative, discussed the ways in which the web has made the world a more connected place for those with disabilities — and yet, how important it is for the people who design digital technologies to be better about making them accessible.
“When the web became available, I could go and travel anywhere,” Brewer said. “There’s a general history of excluding people with disabilities, and then we go and design tech that perpetuates that exclusion. In my vision of the future everything is accessible, including the digital world.”
Liskov captivated the audience with her tales of the early days of computer programming. She was asked to learn Fortran on her first day of work in 1961 — having never written a line of code before.
“I didn’t have any training,” she said. “But then again, nobody did.”
In 1971 Liskov joined MIT, where she created the programming language CLU, which established the notion of “abstract data types” and laid the groundwork for languages like Java and C#. Many coders now take so-called “object-oriented programming” (OOP) for granted: She wryly reflected on how, after she won the Turing Award, one internet commenter looked at her contributions to data abstraction and pointed out that “everybody knows that, anyway.”
“It was a statement to how much the world has changed,” she said with a smile. “When I was doing that work decades earlier, nobody knew anything about [OOP].”
Other researchers built off of Liskov’s remarks in discussing the birth of big data and machine learning. Professor Ronitt Rubinfeld spoke about how computer scientists’ work in sublinear time algorithms has allowed them to better make sense of large amounts of data, while Hamsa Balakrishnan spoke about the ways in which algorithms can help systems engineers make air travel more efficient.
The event’s overarching them was to highlight examples of female role models in a field where they’ve often been overlooked. Paula Hammond, head of MIT’s Department of Chemical Engineering, touted the fact that more than half of undergrads in her department this year were women. Rus urged the women in the audience, many of whom were MIT students, to think about what role they might want to play in continuing to advance science in the coming years.
“To paraphrase our hometown hero, President John F. Kennedy, we need to prepare [women] to see both what technology can do for them — and what they can do for technology,” Rus said.
Rus led the planning of the TEDxMIT event alongside MIT research affiliate John Werner and student directors Stephanie Fu and Rucha Kelkar, both first-years.
MIT announced today that it has created a new associate provost position, to be filled by Timothy Jamison, the Robert R. Taylor Professor of Chemistry and head of the Department of Chemistry. The Institute is also launching an expansive search for a new Institute community and equity officer (ICEO).
The new approach is intended to bolster MIT’s ability to implement programs and strategies that advance diversity, inclusion, equity, a positive climate, and a sense of community. It will also enable the Institute to conduct rigorous self-assessment of its own progress on these issues.
Jamison, who will serve as associate provost for a three-year term, will work with the incoming ICEO to help MIT’s departments create an inclusive campus community. Both Jamison and the ICEO will report to MIT Provost Martin A. Schmidt, who announced the new approach today in an email to the MIT community.
“I am delighted that Tim has agreed to assume this important role. Since 2015, he has led energetic efforts to enhance the quality of life for all members of the Department of Chemistry, and I have been tremendously impressed with his insight, sensitivity, and ability to inspire positive change,” Schmidt wrote in the email.
“I am very grateful for and look forward to this new opportunity to serve the Institute,” Jamison says. “It has been a privilege and pleasure to be head of the Department of Chemistry for the past four years. Looking ahead to this new role, my overarching aim is to support the faculty and their roles in the MIT community. My highest priorities include promoting diversity, inclusion, equity, and community, and to facilitate the search for our next ICEO.”
Alyce Johnson, who has been serving as MIT’s interim ICEO, is retiring this summer after a distinguished career in the Instutute’s leadership ranks. Since last fall, she has been consulting with the MIT community and working with Schmidt to plan the new path forward.
“I am extremely grateful to Alyce for her service as interim ICEO, and for her thoughtful engagement and guidance,” Schmidt wrote to the community.
“I appreciate the broad strategic approach these two roles embody in MIT’s long-standing pursuit of excellence in equity, inclusion and belonging,” Johnson says. “While we continue to collaborate and make forward strides, having dedicated leadership in this area will have a substantial impact on advancing our vision in a more directed and measurable way. We will benefit from the depth of knowledge and experience that both Tim and the new ICEO can bring.”
The new ICEO search will be open to candidates beyond the ranks of MIT faculty, a shift from how the position was originally implemented. This allows MIT to broaden the search and include experts with professional backgrounds in diversity and equity issues. This change was made after consideration of input from the MIT community.
MIT’s ICEO position was created in 2013 to advance activities and public discussion in the areas of community, equity, inclusion, and diversity — comprehensively across the Institute, for students, staff, and faculty. The first ICEO at the Institute, Ed Bertschinger, served from 2013 to 2018 and oversaw a widely read 2015 report identifying a range of inclusion issues in need of ongoing attention.
Jamison will assume his new role beginning July 1. Jamison has been an MIT faculty member since 1999; he earned tenure in 2006 and was promoted to full professor in 2009.
As the new associate provost, Jamison will work to further codify and implement equitable practices across the full range of faculty experiences — including hiring practices, as well as review, promotion, and tenure cases. He says there are also important equity issues centered around the fair distribution of service roles among faculty, which he expects to evaluate as well.
The associate provost will work extensively with MIT’s MindHandHeart coalition — a campus initiative founded in 2015 that develops new approaches in support of health, well-being, and inclusion for people in the MIT community.
MindHandHeart often develops programs tailored to specific portions of the MIT community, an effort that converges with the associate provost’s goal of providing more departmental-level support at MIT, says Maryanne Kirkbride, the executive administrator of MindHandHeart. “We’re looking forward to working with Tim and the next ICEO to develop better individualized support for our academic departments,” she says.
Additionally, Jamison will bring new support to departments, as well as MIT’s five schools and the new MIT Stephen A. Schwarzman College of Computing, to help them create a fully professional climate of inclusion and community in daily life at the Institute.
Jamison brings a record of service and experience to these matters. He and Paula Hammond, head of MIT’s Department of Chemical Engineering and the David H. Koch Professor in Engineering, are currently co-chairs of a working group focused on implementing recommendations from a recent report on sexual harrassment produced by the National Academies of Sciences, Engineering, and Medicine.
The chemistry department, under the supervision of Jamison along with Sarah Rankin, the Institute’s Title IX cooordinator, and Kelley Adams, assistant dean in the Division of Student Life, has also instituted all-inclusive workshops on preventing sexual harrassment at MIT. Similar programs are now being implemented elsewhere at the Institute, including the chemical engineering department.
In the near future, Schmidt stated, he hopes that the presence of Jamison as associate provost, alongside the incoming ICEO, “will help us to move together toward our goal of One MIT.”
Outdated wall art has been replaced with a whiteboard for ideas, couches with an ergonomic work bench, and an old coffee table has made way for a 3-D printer. This is the new craft studio at McCormick Hall that has transformed a previously under-to-unused room into a thriving studio for crafts lovers.
Creating and crafting has long been a tradition at McCormick, MIT’s only all-women and women-identifying dormitory. To honor this tradition, McCormick has had a sewing room since 1967, although its drab ambience and lack of organization had dampened its usage.
“The original sewing room was basically the size of a closet and it was dark, unused, and really cluttered. We wanted a nicer space that more people would be able to use,” says sophomore Nyssa Miller, a resident and chair of sewing at McCormick.
In order to establish a more creative and welcoming space, Nyssa approached Emma Johnson, area director at McCormick, and Lily Gabaree, learning designer at the Media Lab. They brainstormed with other residents of the dorm who showed interest in having a community space to create. With the support of the residents, Johnson and Gabaree applied to the MindHandHeart Innovation Fund and were awarded a grant to found a modern craft studio at McCormick.
“We were really lucky. The process actually went very smoothly. We asked students more about their interests and heard a lot of interest in crafting, 3-D printing, and fiber arts. We talked to the [staff] in the Women’s and Gender Studies Program, and they were really supportive and gave us some ideas about things that were happening on campus. We wrote the proposal for MindHandHeart, which was a great process, and we got funding,” says Gabaree.
After much renovation with the help of MIT Housing and Residential Services, the craft studio opened its doors to McCormick residents and their friends earlier this semester. The entire process, from planning to execution, was an exercise in community building.
The residents of the hall spent several nights assembling furniture from IKEA and organizing an array of crafting tools, including a sewing machine, a serger for advanced sewing, a button maker, a 3-D printer, and other essential supplies for embroidery, knitting, crochet, and woodworking.
One of the first big community projects undertaken in the studio was the creation of McCormick’s "next-generation quilt." A similar quilt was first designed by residents on McCormick’s 50th anniversary six years ago to showcase the diverse ethnicities and cultures of the hall. It is now on display in the hall’s west tower.
“The McCormick Craft Studio is the fantastic result of a community effort … The students have been enthusiastically enjoying the new space and all the cool tools available,” says Raul Radovitzky, professor in the Department of Aeronautics and Astronautics at MIT and head of house of McCormick.
The studio’s founders believe the space encourages women and women-identifying students to continue being creative outside of their academic and work lives. Having an in-dorm space, they attest, will help to foster social connections and reduce isolation.
“Because we are MIT, we are known as hackers and makers, and having that in our dorm actually helps propel the culture that we want as MIT students,” says first-year student Varnika Sinha, a resident in charge of the 3-D printer who conducts regular trainings to instruct residents in the technology.
Now that there is a dedicated space for crafts, Johnson and Gabaree plan to organize open craft nights and more hands-on workshops to engage McCormick’s vibrant community of makers.
Researchers from MIT and elsewhere have developed an interactive tool that, for the first time, lets users see and control how automated machine-learning systems work. The aim is to build confidence in these systems and find ways to improve them.
Designing a machine-learning model for a certain task — such as image classification, disease diagnoses, and stock market prediction — is an arduous, time-consuming process. Experts first choose from among many different algorithms to build the model around. Then, they manually tweak “hyperparameters” — which determine the model’s overall structure — before the model starts training.
Recently developed automated machine-learning (AutoML) systems iteratively test and modify algorithms and those hyperparameters, and select the best-suited models. But the systems operate as “black boxes,” meaning their selection techniques are hidden from users. Therefore, users may not trust the results and can find it difficult to tailor the systems to their search needs.
In a paper presented at the ACM CHI Conference on Human Factors in Computing Systems, researchers from MIT, the Hong Kong University of Science and Technology (HKUST), and Zhejiang University describe a tool that puts the analyses and control of AutoML methods into users’ hands. Called ATMSeer, the tool takes as input an AutoML system, a dataset, and some information about a user’s task. Then, it visualizes the search process in a user-friendly interface, which presents in-depth information on the models’ performance.
“We let users pick and see how the AutoML systems works,” says co-author Kalyan Veeramachaneni, a principal research scientist in the MIT Laboratory for Information and Decision Systems (LIDS), who leads the Data to AI group. “You might simply choose the top-performing model, or you might have other considerations or use domain expertise to guide the system to search for some models over others.”
In case studies with science graduate students, who were AutoML novices, the researchers found about 85 percent of participants who used ATMSeer were confident in the models selected by the system. Nearly all participants said using the tool made them comfortable enough to use AutoML systems in the future.
“We found people were more likely to use AutoML as a result of opening up that black box and seeing and controlling how the system operates,” says Micah Smith, a graduate student in the Department of Electrical Engineering and Computer Science (EECS) and a researcher in LIDS.
“Data visualization is an effective approach toward better collaboration between humans and machines. ATMSeer exemplifies this idea,” says lead author Qianwen Wang of HKUST. “ATMSeer will mostly benefit machine-learning practitioners, regardless of their domain, [who] have a certain level of expertise. It can relieve the pain of manually selecting machine-learning algorithms and tuning hyperparameters.”
Joining Smith, Veeramachaneni, and Wang on the paper are: Yao Ming, Qiaomu Shen, Dongyu Liu, and Huamin Qu, all of HKUST; and Zhihua Jin of Zhejiang University.
Tuning the model
At the core of the new tool is a custom AutoML system, called “Auto-Tuned Models” (ATM), developed by Veeramachaneni and other researchers in 2017. Unlike traditional AutoML systems, ATM fully catalogues all search results as it tries to fit models to data.
ATM takes as input any dataset and an encoded prediction task. The system randomly selects an algorithm class — such as neural networks, decision trees, random forest, and logistic regression — and the model’s hyperparameters, such as the size of a decision tree or the number of neural network layers.
Then, the system runs the model against the dataset, iteratively tunes the hyperparameters, and measures performance. It uses what it has learned about that model’s performance to select another model, and so on. In the end, the system outputs several top-performing models for a task.
The trick is that each model can essentially be treated as one data point with a few variables: algorithm, hyperparameters, and performance. Building on that work, the researchers designed a system that plots the data points and variables on designated graphs and charts. From there, they developed a separate technique that also lets them reconfigure that data in real time. “The trick is that, with these tools, anything you can visualize, you can also modify,” Smith says.
Similar visualization tools are tailored toward analyzing only one specific machine-learning model, and allow limited customization of the search space. “Therefore, they offer limited support for the AutoML process, in which the configurations of many searched models need to be analyzed,” Wang says. “In contrast, ATMSeer supports the analysis of machine-learning models generated with various algorithms.”
User control and confidence
ATMSeer’s interface consists of three parts. A control panel allows users to upload datasets and an AutoML system, and start or pause the search process. Below that is an overview panel that shows basic statistics — such as the number of algorithms and hyperparameters searched — and a “leaderboard” of top-performing models in descending order. “This might be the view you’re most interested in if you’re not an expert diving into the nitty gritty details,” Veeramachaneni says.
Similar visualization tools present this basic information, but without customization capabilities. ATMSeer includes an “AutoML Profiler,” with panels containing in-depth information about the algorithms and hyperparameters, which can all be adjusted. One panel represents all algorithm classes as histograms — a bar chart that shows the distribution of the algorithm’s performance scores, on a scale of 0 to 10, depending on their hyperparameters. A separate panel displays scatter plots that visualize the tradeoffs in performance for different hyperparameters and algorithm classes.
Case studies with machine-learning experts, who had no AutoML experience, revealed that user control does help improve the performance and efficiency of AutoML selection. User studies with 13 graduate students in diverse scientific fields — such as biology and finance — were also revealing. Results indicate three major factors — number of algorithms searched, system runtime, and finding the top-performing model — determined how users customized their AutoML searches. That information can be used to tailor the systems to users, the researchers say.
“We are just starting to see the beginning of the different ways people use these systems and make selections,” Veeramachaneni says. “That’s because now that this information is all in one place, and people can see what’s going on behind the scenes and have the power to control it.”
In the fall of 2018, three first-year students at Trinity University in Texas had an idea for a nonprofit that would connect artists with the special-needs community and help people engage with artwork regardless of their age or disability.
Fortunately for the students, Trinity had just launched a new program on campus to support entrepreneurs. In November, the students began meeting regularly with a team of experienced mentors from their community. The mentors helped the students refine their idea and prioritize their next steps to form an organization. The following semester, the project, called heARTful, was named a finalist in a local venture competition and awarded $5,000.
Around the same time, a family in Mobile, Alabama, was looking to adapt its furniture store to the world of online retail. It became a member of the first cohort of companies to go through a team-based mentoring program hosted by the University of South Alabama. The program also included a dental practices management platform, an online retailer of cigars, and someone with a software solution to help coordinate fishing tournaments that has raised almost $1 million to date.
What do all of these ventures have in common? They’re all benefiting from a mentoring model for entrepreneurs that was developed at MIT nearly 20 years ago.
The core methodology of MIT’s Venture Mentoring Service (VMS) is straightforward: Entrepreneurs meet with a team of mentors in ongoing, confidential meetings. The mentors are volunteers and commit to avoiding any conflict of interest to ensure that they give objective and unbiased advice.
But while the basic tenets of the model are simple, they require a robust and disciplined support structure to be effective.
The VMS team, facing increasing interest from outside organizations, launched the VMS Outreach Training Program in 2006 to formally disseminate the model to other organizations.
The Outreach Training Program has since trained close to 100 organizations around the world, including economic development organizations, business accelerators and incubators, and around 40 colleges and universities.
In that time, the VMS model has been tested on campuses of all types, and its success has earned it a reputation in higher education as an outstanding methodology for supporting entrepreneurs.
A model for impact
In 2000, VMS was founded by the late MIT Professor David Staelin and the late Alexander Dingee ’52, both successful serial entrepreneurs. Not much has changed from their early idea: Members of the MIT community, including students, faculty, staff, and alumni, can be at any stage of venture creation when they begin using VMS. A team of three to five volunteer mentors, meeting in person with the entrepreneur, provides business advice through a carefully structured process.
The program attracted attention from MIT’s peer institutions almost immediately.
“As we talk about our program with other organizaions, the concept of the trusted environment for the entrepreneur resonates; many people see that value,” says VMS Outreach Training Program Manager Ariane Martins, who also believes the team mentoring approach is a key to the program’s success. “Most people we talk to have never done mentoring in teams — certainly not in this structured way — but what we’ve seen is that it raises the quality and breadth of advice the entrepreneurs are given.”
The Outreach Training Program was formed in response to a growing demand, which was driven in part by several long-term trends in higher education, according to VMS Co-director Jerome Smith.
“These days, there are very few universities that aren’t talking about entrepreneurship or innovation” Smith says. “[VMS] is very complementary to other programs, because it is very practical. People gain a lot from the academic theory of entrepreneurship, but mentors in the VMS Model are working with entrepreneurs who are actually trying to start or grow a business.”
Schools have also collaborated with VMS to address increasing student interest in entrepreneurship. Luis Martinez is the director of the Center for Innovation and Entrepreneurship at Trinity University. Having worked with students in higher education for the better part of the last 20 years, he jokingly refers to college students today as “the Shark Tank generation.”
“When I was growing up, it used to be cool to start a band,” Martinez says. “Now it’s cool to start a company.”
As the VMS methodology has spread, its applicability has been proven in a wide range of settings, from Mexico to Australia. Some campuses, like Trinity, a small, liberal arts school, predominantly serve undergraduates with VMS, while others primarily help professors, researchers, and even members of the local community.
The University of Texas (UT) has participated in the Outreach Training Program three times to implement the VMS model on a number of campuses including the UT MD Anderson Cancer Center and UT Austin.
“Texas is very, very different from Boston, but what we’ve found is that the core principals of team-based, conflict-free mentoring still hold true everywhere we’ve tested it,” says Matt Sorenson, innovation program manager for the University of Texas system. “Each ecosystem is so different, but we’ve found the methodology is equally effective.”
Universities lifting communities
University officials also say the experiences of both the mentee and mentors in the VMS program can make a difference in the communities around their campuses.
“The impact [of the Trinity VMS program] has already been made in the community,” Martinez says. “It’s exciting, there’s a whole group of people now trained in the VMS methodology, both in the programming and the mentoring, so we’re trying to leverage the lessons being learned around the state and city.”
When the University of South Alabama decided to adopt the VMS methodology, it partnered with the city of Mobile, the local county, and the Chamber of Commerce to offer mentoring services not just to people affiliated with the school but also to local business owners.
“What’s great for smaller communities like ours is all this [meeting and learning in groups] means you’re building infrastructure,” says Michael Chambers, the associate vice president of research at the University of South Alabama. “All of a sudden you have an organized network of mentors, and they become aware of all these other local companies, and they become cheerleaders for those companies when they’re out in the community. … I don’t know where else we’d get that.”
On a sunny May day, Pierre-Luc Vautrey sits in 1369 Coffeehouse in Cambridge, talking enthusiastically about his work — five research projects to be exact. He speaks quickly, and the coffee gives him an extra boost. He has a lot of ground to cover, and at times he has to re-explain certain areas of his research. Luckily, he’s patient and wants to ensure that people understand his work.
Vautrey is a third-year doctoral student in MIT’s Department of Economics. While he spent his undergraduate years studying applied math and physics in his home country of France, he was always drawn to the humanities and social sciences.
“I still had this itch to go back to social science at some point. It just seemed like a really nice way to bridge science and quantitative approach with social science and humans in general. That’s how I got into economics,” he says.
As a behavioral economist, Vautrey aims to extend our understanding of economic decisions using psychology. This approach questions traditional assumptions, ever so slightly, in order to make outcomes more realistic regarding human behavior.
“Traditional economics has been modeling everything as rational. We assume that the agent learns like a statistician and makes rational decisions. And in the last 20 or 30 years, this model has shown its limits. It’s still very popular for many things, but for others we can do a lot better at explaining people’s behavior and why certain social systems work and some systems don’t work, by using psychology [to understand] how people actually think and make decisions,” he says.
The unifying theme throughout his current work is understanding how people form beliefs and expectations.
“You can use psychology to take a small departure, that’s the key, from rational behavior, which is having correct expectations and basing decisions on these expectations,” he says. “You still make decisions based on expectations, but you have incorrect beliefs for various psychological reasons. That’s kind of the key psychological, irrational approach that I’m interested in. What is the role of beliefs, how do we best measure them, and in various contexts can we explain why people have irrational beliefs? Can we predict incorrect beliefs of people based on context? Does it help us explain sometimes puzzling decisions?”
One of the projects Vautrey is working on, along with Professor Frank Schilbach from the Department of Economics, is how mental health affects beliefs and economic decision making. They began conducting research in India among people with depression in low-income communities with no access to mental health services. They want to evaluate whether depression affects a person’s self-confidence and, consequently, their ability to participate in their economy. They are working with Sangath, an NGO providing low-cost psychotherapy to the study’s participants, to measure the effects of psychotherapy on not only mental health, but also economic decisions. Vautrey began working on the project the fall of 2017, during its early brainstorming stages, and has visited India twice since the field work began.
“You have to go there to see how operations are going, see the actual participants, because it's really hard to get everything from calls. You have people implementing the project, but usually the people who have designed the questions or are initiating the idea are not full-time in the field because they are professors so they have to teach,” Vautrey explains.
Field visits are also important in order to see whether the research objective and the information gathered are consistent with each other.
“You have to design questions that are qualitative, that are verbal, but are going to generate numerical outcomes that you can analyze. It’s a back-and-forth between sociological-style research, when you talk to people and try to understand what they think, and how you go from there to build quantitative measures. You have to be on the field; you have to be face-to-face to understand whether your numeric outcome is consistent with what you want it to mean,” he says.
Traveling is important to executing research, and Vautrey enjoys that aspect of the job. He has loved traveling since his youth and has taken as many opportunities as he could to do so.
Beyond the project in India, Vautrey is working on a few other projects, two more in progress and two in their preliminary stages. In the former two, he is studying how people choose biased information sources and how people are influenced by news repetition. In another project with MIT economics doctoral student Charlie Rafkin, Vautrey is investigating unsafe driving patterns in developing countries and how drivers’ motivated reasoning about road safety leads to more risk taking that could be easily avoided by correcting drivers’ beliefs and overconfidence.
Vautrey’s newest endeavor is taking him to Colombia with Pedro Bessone Tepedino, another MIT economics doctoral student, for preliminary research for a new project centered around crime and teenage involvement in gangs.
While he enjoys doing all of his research, Vautrey finds that the work can make life a bit unstructured at times. He grounds himself by staying active with activities such as biking and rock climbing.
In the future, Vautrey hopes to work in academia. As a professor, he isn’t sure what specifically he wants to specialize in quite yet, but he says that it will likely have something to do with using psychology and economics to answer specific questions linked to poverty and development. He found a love for teaching through his work as a teaching assistant at MIT this past semester. It requires patience, but Vautrey finds the work rewarding.
“It’s a really nice feeling when you manage to get someone to understand something you said. When you have a class, it’s almost impossible to get everyone to understand everything you want,” he says, adding, “To me, if I get half of the class to understand something and to learn something they really value, I’m already happy.”
With a couple billion more people estimated to join the global population in the next few decades, world food production could use an upgrade. Africa has a key role to play: Agriculture is Africa’s biggest industry, but much of Africa’s agricultural land is currently underutilized. Crop yields could be increased with more efficient farming techniques and new equipment — but that would require investment capital, which is often an obstacle for farmers.
A new research collaboration at the MIT Institute for Data, Systems, and Society (IDSS) aims to address this challenge with data. The group plans to use data from technologically advanced farms to better predict the value of intervention in underperforming farms. Ultimately, the goal is to create a platform for sharing data and risk among invested parties, from farmers and lenders to insurers and equipment manufacturers.
Sharing data, sharing risk
Many African farmers lack the capital to invest in yield-increasing upgrades like new irrigation systems, new machinery, new fertilizers, and technology for sensing and tracking crop growth. The most common path to capital is bank loans, with land as collateral. This is an unattractive proposition for farmers, who already bear the many risks of production, including bad weather, changing market prices, or even the shocks of geopolitical events.
Lenders, on the other hand, have an incomplete assessment of their risk, especially with potential borrowers who have no credit history. Lenders also lack data and tools to predict their return on investment.
“Building a platform for risk-sharing is key to upgrading farming practices,” says Munther Dahleh, a professor of electrical engineering and computer science at MIT and director of IDSS. In order to create such a platform, Dahleh and the IDSS team aim to better predict the value of employing advanced farming practices on the production of individual farms. This prediction needs to be accurate enough to incentivize investment from economic stakeholders and the farmers themselves, who are in competition with each other and may be reluctant to share information.
The IDSS approach proposes a data-sharing platform that incentivizes all parties to participate: Technologically advanced farms are rewarded for their valuable data, bankers benefit from data that support their credit risk models, farmers get better loan terms and recommendations that increase their profits and production, and technology companies get recommendations on how to best support the needs of their farmer customers. “Such a platform has to have the correct incentives to engage everyone to participate, have sufficient protection from players with market power, and ultimately provide valuable data for farmers and creditors alike,” says Dahleh.
The absence of data from underperforming farms presents a challenge to extrapolating the value of intervention and assessing the uncertainty in such predictions. With sparse available data, researchers are looking to conduct experiments in strategically selected farms to provide valuable new data for the rest. Researchers will use advanced machine learning, including active learning methodology, to try to achieve both a quantification of the predicted value of intervention and a quantification of the uncertainty of that prediction to a degree of confidence. Once more data is available, IDSS researchers intend to refine their calculations and develop new techniques for extrapolating the value of intervention in less-advanced farms.
One likely intervention for many African farmers involves using different fertilizers. Many farmers aren’t currently using fertilizers targeted to specific soil or various stages of farming — so fertilizer producers are another vested interest in this agriculture economy.
To help these farmers get access to better loan terms, Moroccan phosphate company OCP is funding a collaboration between IDSS researchers and Mohammed VI Polytechnic University (UM6P) in Morocco. This research collaboration with OCP, a leading global company in the phosphate fertilizer industry, includes building the data- and risk-sharing platform as well as other foundational research in agriculture. The collaboration has the potential to engage other stakeholders working or investing in African agriculture.
“This collaboration will help accelerate our efforts to develop pertinent solutions for African agriculture using high-level agri-tech tools,” says Fassil Kebede, professor of soil science and head of the Center for Soil and Fertilizer Research in Africa. “This will offer farmers possibilities for better production and growth, which is part of our mission to contribute to Africa’s food-security objectives.”
“African farmers are at the heart of the OCP Group’s mission and strategy, while data analytics and predictive tools are today essential for agriculture development in Africa,” adds Mostafa Terrab, OCP Group chair and CEO. “This collaboration with IDSS will help us bring together new technology and analytical methods from one side, and our expertise with African farmers and their challenges from the other side. It will reinforce our capabilities to offer adapted solutions to African farmers, especially small holders, to enable them to make more precise and timely decisions.”
Ultimately, IDSS aims to bring wins across an entire economic ecosystem, from insurers to lenders to equipment and fertilizer companies. But most importantly, boosting this ecosystem could help lift many farmers out of poverty — and bring about a much-needed increase in the world’s aggregate food production.
Says Dahleh: “To accomplish this mission, this project will demonstrate the power of data coupled with advanced tools from predictive analytics, machine learning, reinforcement learning, and data sharing markets.”
If you’ve ever wondered what a loaf of bread would look like as a cat, edges2cats is for you. The program that turns sketches into images of cats is one of many whimsical creations inspired by Phillip Isola’s image-to-image translation software released in the early days of generative adversarial networks, or GANs. In a 2016 paper, Isola and his colleagues showed how a new type of GAN could transform a hand-drawn shoe into its fashion-photo equivalent, or turn an aerial photo into a grayscale map. Later, the researchers showed how landscape photos could be reimagined in the impressionist brushstrokes of Monet or Van Gogh. Now an assistant professor in MIT’s Department of Electrical Engineering and Computer Science, Isola continues to explore what GANs can do.
GANs work by pairing two neural networks, trained on a large set of images. One network, the generator, outputs an image patterned after the training examples. The other network, the discriminator, rates how well the generator’s output image resembles the training data. If the discriminator can tell it’s a fake, the generator tries again and again until its output images are indistinguishable from the examples. When Isola first heard of GANs, he was experimenting with nearest-neighbor algorithms to try to infer the underlying structure of objects and scenes.
GANs have an uncanny ability to get at the essential structure of a place, face, or object, making structured prediction easier. Introduced five years ago, GANs have been used to visualize the ravages of climate change, produce more realistic computer simulations, and protect sensitive data, among other applications.
To connect the growing number of GAN enthusiasts at MIT and beyond, Isola has recently helped to organize GANocracy, a day of talks, tutorials, and posters being held at MIT on May 31 that is co-sponsored by the MIT Quest for Intelligence and MIT-IBM Watson AI Lab. Isola recently spoke about the future of GANs.
Q: Your image-to-image translation paper has more than 2,000 citations. What made it so popular?
A: It was one of the earlier papers to show that GANs are useful for predicting visual data. We showed that this setting is very general, and can be thought of as translating between different visualizations of the world, which we called image-to-image translation. GANs were originally proposed as a model for producing realistic images from scratch. But the most useful application may be structured prediction, which is what GANs are mostly being used for these days.
Q: GANs are easily customized and shared on social media. Any favorites among these projects?
A: #Edges2cats is probably my favorite, and it helped to popularize the framework early on. Architect Nono Martínez Alonso has used pix2pix for exploring interesting tools for sketch-based design. I like everything by Mario Klingemann; Alternative Face is especially thought-provoking. It puts one person’s words into someone else’s mouth, hinting at a potential future of “alternative facts.” Scott Eaton is pushing the limits of GANs by translating sketches into 3-D sculptures.
Q: What other GAN art grabs you?
A: I really like all of it. One remarkable example is GANbreeder. It’s a human-curated evolution of GAN-generated images. The crowd chooses which images to breed or kill off. Over many generations, we end up with beautiful and unexpected images.
Q: How are GANs being used beyond art?
A: In medical imaging, they’re being used to generate CT scans from MRIs. There’s potential there, but it can be easy to misinterpret the results: GANs are making predictions, not revealing the truth. We don't yet have good ways to measure the uncertainty of their predictions. I'm also excited about the use of GANs for simulations. Robots are often trained in simulators to reduce costs, creating complications when we deploy them in the real world. GANs can help bridge the gap between simulation and reality.
Q: Will GANs redefine what it means to be an artist?
A: I don't know, but it's a super-interesting question. Several of our GANocracy speakers are artists, and I hope will touch on this. GANs and other generative models are different than other kinds of algorithmic art. They are trained to imitate, so the people being imitated probably deserve some credit. The art collective, Obvious, recently sold a GAN image at Christie's for $432,500. Obvious selected the image, signed and framed it, but the code was derived from work by then-17-year-old Robbie Barrat. Ian Goodfellow helped develop the underlying algorithm.
Q: Where is the field heading?
A: As amazing as GANs are, they are just one type of generative model. GANs might eventually fade in popularity, but generative models are here to stay. As models of high-dimensional structured data, generative models get close to what we mean when we say “create,” “visualize,” and “imagine.” I think they will be used more and more to approximate capabilities that still seem uniquely human. But GANs do have some unique properties. For one, they solve the generative modeling problem via a two-player competition, creating a generator-discriminator arms race that leads to emergent complexity. Arms races show up across machine learning, including in the AI that achieved superhuman abilities in the game Go.
Q: Are you worried about the potential abuse of GANs?
A: I’m definitely concerned about the use of GANs to generate and spread misleading content, or so-called fake news. GANs make it a lot easier to create doctored photos and videos, where you no longer have to be a video editing expert to make it look like a politician is saying something they never actually said.
Q: You and the other GANocracy organizers are advocating for so-called GANtidotes. Why?
A: We would like to inoculate society against the misuse of GANs. Everyone could just stop trusting what we see online, but then we’d risk losing touch with reality. I’d like to preserve a future in which “seeing is believing.” Luckily, many people are working on technical antidotes that range from detectors that seek out the telltale artifacts in a GAN-manipulated image to cryptographic signatures that verify that a photo has not been edited since it was taken. There are a lot of ideas out there, so I’m optimistic it can be solved.
MIT professors Senthil Todadri and Xiao-Gang Wen are members of the newly established Simons Collaboration on Ultra-Quantum Matter. The effort, funded by the Simons Foundation, is an $8 million four-year award, renewable for three additional years, and will support theoretical physics research across 12 institutions, including MIT.
The science of the collaboration is based on a series of recent developments in theoretical physics, revealing that even large macroscopic systems that consist of many atoms or electrons — matter — can behave in an essentially quantum way. Such ultra-quantum matter (UQM) allows for quantum phenomena beyond what can be realized by individual atoms or electrons, including distributed storage of quantum information, fractional quantum numbers, and perfect conducting boundary.
While some examples of UQM have been experimentally established, many more have been theoretically proposed, ranging from highly entangled topological states to unconventional metals that behave like a complex soup. The Simons Collaboration on Ultra-Quantum Matter will classify possible forms of UQM, understand their physical properties, and provide the key ideas to enable new realizations of UQM in the lab.
Ultra dream team
In particular, the collaboration will draw upon lessons from recently discovered connections between topological states of matter and unconventional metals, and seeks to develop a new theoretical framework for those phases of ultra-quantum matter. Achieving these goals requires ideas and tools from multiple areas of theoretical physics, and accordingly the collaboration brings together experts in condensed matter physics, quantum field theory, quantum information, and atomic physics to forge a new interdisciplinary approach.
Directed by Professor Ashvin Vishwanath at Harvard University, the collaboration comprises researchers at MIT, Harvard, Caltech, the Institute for Advanced Study, Stanford University, University of California at Santa Barbara, University of California at San Diego, University of Chicago, University of Colorado at Boulder, University of Innsbruck, University of Maryland, and University of Washington.
“I am looking forward to scientific interactions with MIT theorists Senthil and Wen, who are key members of our Simons collaboration on Ultra-Quantum Matter, and hope this will further strengthen collaborations within the Cambridge area and beyond. Their research on highly entangled quantum materials is of fundamental significance, and may provide new directions for device applications, quantum computing, and high-temperature superconductors,” says collaboration director Ashvin Vishwanath of Harvard University.
“They have also been mentors for several collaboration members,” says Vishwanath, who worked with Senthil as a Pappalardo Fellow in physics from 2001 to 2004.
Senthil has played a leading role in the field of non-Fermi liquids, in the classification of strongly interacting topological insulators and related topological phases, and in the development of field theory dualities with diverse applications in condensed matter physics.
Wen is one of the founders of the field of topological phases of matter, introducing the concept of topological order in 1989 and opening up a new research direction in condensed matter physics. Wen’s research has often exposed mathematical structures that have not appeared before in condensed matter physics problems.
Of the 17 faculty members who are participating in the collaboration, more than half, including Senthil, Wen, and Vishwanath, have MIT affiliations.
Michael Hermele, the collaboration’s deputy director and an associate professor at the University of Colorado at Boulder, was a postdoc in the MIT Condensed Matter Theory group.
Associate professors Xie Chen PhD ’12 and Michael Levin PhD ’06, at Caltech and the University of Chicago, respectively, earned their doctorates at MIT under Wen.
Other principal investigators include alumni Subir Sachdev ’82, now chair of the Department of Physics at Harvard, and Leon Balents ’89, a physics professor at UC Santa Barbara's Kavli Institute for Theoretical Physics. John McGreevy, a string theorist who conducted research in the Center for Theoretical Physics (CTP), is now a professor of physics at UC San Diego. Dam Thanh Son and Andreas Karch, former CTP postdocs, are now with the University of Chicago and the University of Washington, respectively.
The collaboration is part of the Simons Collaborations in Mathematics and Physical Sciences program, which aims to “stimulate progress on fundamental scientific questions of major importance in mathematics, theoretical physics and theoretical computer science.” The Simons Collaboration on Ultra-Quantum Matter is one of 12 such collaborative grants ranging across these fields.
The first meeting of the newly established collaboration will take place Sept. 12-13 in Cambridge, Massachusetts.
Agricultural productivity technologies for small-holder farmers; food safety solutions for everyday consumers; sustainable supply chain interventions in the palm oil industry; water purification methods filtering dangerous micropollutants from industrial and wastewater streams — these are just a few of the research-based solutions being supported by the Abdul Latif Jameel Water and Food Systems Lab (J-WAFS) at MIT. J-WAFS is funding these and other projects through its fifth round of seed grants, providing over $1 million in funding to the MIT research community. These grants, which are funded competitively to MIT principal investigators (PIs) across all five schools at the Institute, exemplify the ambitious goals of MIT’s Institute-wide effort to address global water and food systems challenges through research and innovation.
This year, seven new projects led by nine faculty PIs across all five schools will be funded with two-year grants of up to $150,000, overhead-free. Interest in water and food systems research at MIT is substantial, and growing. By the close of this grant cycle, over 12 percent of MIT faculty will have submitted J-WAFS grant proposals. Thirty-four principal investigators submitted proposals to this latest call, nearly one third of whom were proposing to J-WAFS for the first time. “The broad range of disciplines that this applicant pool represents demonstrates how meeting today’s water and food challenges is motivating many diverse researchers in our community," comments Renee Robins, executive director of J-WAFS. "Our reach across all of MIT’s schools further attests to the strength of the Institute’s capabilities that can be applied to the search for solutions to pressing water and food sector challenges.” The nine faculty who were funded represent eight departments and labs, including the departments of Civil and Environmental Engineering, Mechanical Engineering, Chemical Engineering, Chemistry, and Economics, as well as the Media Lab (School of Architecture and Planning), MIT D-Lab (Office of the Vice Chancellor), and the Sloan School of Management.
New approaches to ensure safe drinking water
Nearly 1 billion people worldwide receive their drinking water through underground pipes that only operate intermittently. In contrast to continuous water supplies, pipes like these that are only filled with water during limited supply periods are vulnerable to contamination. However, it is challenging to quantify the quality of water that comes out of these pipes because of the vast differences in how the pipe networks are arranged and where they are located, especially in dense urban settings. Andrew J. Whittle, the Edmund K. Turner Professor in Civil Engineering, seeks to address this problem by gathering and making available more precise data on how water quality is affected by how the pipe is used — i.e., during periods of filling, flushing, or stagnation. Supported by the seed grant, he and his research team will perform tests in a section of abandoned pipe in Singapore, one that is still connected to the urban water pipe network there. By controlling flushing rates, monitoring stagnation, and measuring contamination, the study will analyze how variances in flow affect water quality, and evaluate how these data might be able to inform future water quality studies in cities with similar piped water challenges.
Patrick Doyle, the Robert T. Haslam (1911) Professor of Chemical Engineering, is taking a different approach to water quality: creating a filter to remove micropollutants. Wastewater from industrial and agricultural processes often contains solvents, petrochemicals, lubricants, pharmaceuticals, hormones, and pesticides, which can enter natural water systems. While these micropollutants may be present at low concentrations, they can still have a significant negative impact on aquatic ecosystems, as well as human health. The challenge is in detecting and removing these micropollutants, because of the low concentrations in which they occur. For this project, Doyle and his team will develop a system to remove a variety of micropollutants, at even the smallest concentrations, using a special hydrogel particle that can be “tuned” to fit the size and shape of particular particles. Leveraging the flexibility of these hydrogels, this technology can improve the speed, precision, efficiency, and environmental sustainability of industrial water purification systems, and improve the health of the natural water systems upon which humans and our surrounding ecosystems rely.
Developing support tools for small-holder farmers
More than half of food calories consumed globally — and 70 percent of food calories consumed in developing countries — are supplied by approximately 475 million small-holder households in developing and emerging economies. These farmers typically operate through informal contracts and processes, which can lead to large economic inefficiencies and lack of traceability in the supply chains that they are a part of. Joann de Zegher, the Maurice F. Strong Career Development Professor in the operations management program at the MIT Sloan School of Management, seeks to address these challenges by developing a mobile-based trading platform that links small-holder farmers, middlemen, and mills in the palm oil supply chain in Indonesia. Rapid growth in demand in this industry has led to high environmental costs, and recently pressure from consumers and nongovernmental organizations is motivating producers to employ more sustainable practices. However, these pressures deepen market access challenges for small-holder palm oil farmers. Her project seeks to improve the efficiency and effectiveness of the current supply chain, and create transparency as a byproduct.
Another small-holder farmer intervention is being developed by Robert M. Townsend, the Elizabeth and James Killian Professor of Economics. He is leading a research effort to improve access to crop insurance for small-holder farmers, who are particularly vulnerable to weather-related crop failures. Crop cultivation worldwide is highly vulnerable to unfavorable weather. In developing countries, farmers bear the financial burden of their crops’ exposure to weather ravages, the extent of which will only increase due to the effects of climate change. As a result, they rely on low-risk, low-yield cultivation practices that do not allow for the food and financial gains that can be possible when favorable weather supports higher yields. While crop insurance can help, it is often prohibitively expensive for these small-scale producers. Townsend and his research team seek to make crop insurance more accessible and affordable for farmers in developing regions by developing a new system of insurance pricing and payoff schedules that takes into account the widely varying ways through which weather affects crop’s development and yield throughout the growth cycle. Their goal is to provide a new, personalized insurance tool that improves farmers’ ability to protect their yields, invest in their crops, and adapt to climate change in order to stabilize food supply and farmer livelihoods worldwide.
Access to affordable fertilizer is another challenge that small holders face. Ammonia is the key ingredient in fertilizers; however, most of the world’s supply is produced by the Haber-Bosch process, which directly converts nitrogen and hydrogen gas to ammonia in a highly capital-intensive process that is difficult to downscale. Finding an alternative way to synthesize ammonia could transform access to fertilizer and improve food security, particularly in the developing world where current fertilizers are prohibitively expensive. For this seed grant project, Yogesh Surendranath, Paul M Cook Career Development Assistant Professor in the Department of Chemistry, will develop an electrochemical process to synthesize ammonia, one that can be powered using renewable energy sources such as solar or wind. Designed to be implemented in a decentralized way, this technology could enable fertilizer production directly in the fields where it is needed, and would be especially beneficial in developing regions without access to existing ammonia production infrastructure.
Even when crops produce high yields, post-harvest preservation is a challenge, especially to fruit and vegetable farmers on small plots of land in developing regions. The lack of affordable and effective post-harvest vegetable cooling and storage poses a significant challenge for them, and can lead to vegetable spoilage, reduced income, and lost time. Most techniques for cooling and storing vegetables rely on electricity, which is either unaffordable or unavailable for many small-holder farmers, especially those living on less than $3 per day in remote areas. The solution posed by an interdisciplinary team led by Daniel Frey, professor in the Department of Mechanical Engineering and D-Lab faculty director, along with Leon Glicksman, professor of architecture and mechanical engineering, is a storage technology that uses the natural evaporation of water to create a cool and humid environment that prevents rot and dehydration, all without the need for electricity. This system is particularly suited for hot, dry regions such as Kenya, where the research team will be focusing their efforts. The research will be conducted in partnership with researchers from University of Nairobi’s Department of Plant Science and Crop Protection, who have extensive experience working with low-income rural communities on issues related to horticulture and improving livelihoods. The team will build and test evaporative cooling chambers in rural Kenya to optimize the design for performance, practical construction, and user preferences, and will build evidence for funders and implementing organizations to support the dissemination of these systems to improve post-harvest storage challenges.
Combatting food safety challenges through wireless sensors
Food safety is a matter of global concern, and a subject that several J-WAFS-funded researchers seek to tackle with innovative technologies. And for good reason: Food contamination and foodborne pathogens cause sickness and even death, as well as significant economic costs including the wasted labor and resources that occur when a contaminated product is disposed of, the lost profit to affected companies, and the lost food products that could have nourished a number of people. Fadel Adib, an assistant professor at the MIT Media Lab, will receive a seed grant to develop a new tool that quickly and accurately assesses whether a given food product is contaminated. This food safety sensor uses wireless signals to determine the quality and safety of packaged food using a radio-frequency identification sticker placed on the product’s container. The system turns off-the-shelf RFID tags into spectroscopes which, when read, can measure the material contents of a product without the need to open its package. The sensor can also identify the presence of contaminants — pathogens as well as adulterants that affect the nutritional quality of the food product. If successful, this research, and the technology that results, will pave the way for wireless sensing technologies that can inform their users about the health and safety of their food and drink.
With these seven newly funded projects, J-WAFS will have funded 37 total seed research projects since its founding in 2014. These grants serve as important catalysts of new water and food sector research at MIT, resulting in publications, patents, and other significant research support. To date, J-WAFS’ seed grant PIs have been awarded over $11M in follow-on funding. J-WAFS’ director, Professor John Lienhard, commented on the influence of this grant program: “The betterment of society drives our research community at MIT. Water and food, our world’s most vital resources, are currently put at great risk by a variety of global-scale challenges, and MIT researchers are responding forcefully. Through this, and J-WAFS’ other grant programs, we see MIT's creative innovations and actionable solutions that will help to ensure a sustainable future.”
J-WAFS Seed Grants, 2019
PI: Fadel Adib, assistant professor, MIT Media Lab
PI: Joann de Zegher, Maurice F. Strong Career Development Professor, Sloan School of Management
PI: Patrick Doyle, Robert T. Haslam (1911) Professor of Chemical Engineering, Department of Chemical Engineering
PIs: Daniel Frey, professor, Department of Mechanical Engineering, and faculty research director, MIT D-Lab; Leon Glicksman, professor of building technology and mechanical engineering, Department of Mechanical Engineering
PI: Yogesh Surendranath, Paul M Cook Career Development Assistant Professor, Department of Chemistry
- Designing Purely Weather-Contingent Crop Insurance with Personalized Coverage to Improve Farmers’ Investments in their Crops for Higher Yields
PI: Robert M. Townsend, Elizabeth and James Killian Professor of Economics, Department of Economics
PI: Andrew J. Whittle, Edmund K. Turner Professor in Civil Engineering, Department of Civil and Environmental Engineering
Voice assistants like Siri and Alexa can tell the weather and crack a good joke, but any 8-year-old can carry on a better conversation.
The deep learning models that power Siri and Alexa learn to understand our commands by picking out patterns in sequences of words and phrases. Their narrow, statistical understanding of language stands in sharp contrast to our own creative, spontaneous ways of speaking, a skill that starts developing even before we are born, while we're still in the womb.
To give computers some of our innate feel for language, researchers have started training deep learning models on the grammatical rules that most of us grasp intuitively, even if we never learned how to diagram a sentence in school. Grammatical constraints seem to help the models learn faster and perform better, but because neural networks reveal very little about their decision-making process, researchers have struggled to confirm that the gains are due to the grammar, and not the models’ expert ability at finding patterns in sequences of words.
Now psycholinguists have stepped in to help. To peer inside the models, researchers have taken psycholinguistic tests originally developed to study human language understanding and adapted them to probe what neural networks know about language. In a pair of papers to be presented in June at the North American Chapter of the Association for Computational Linguistics conference, researchers from MIT, Harvard University, University of California, IBM Research, and Kyoto University have devised a set of tests to tease out the models’ knowledge of specific grammatical rules. They find evidence that grammar-enriched deep learning models comprehend some fairly sophisticated rules, performing better than models trained on little-to-no grammar, and using a fraction of the data.
“Grammar helps the model behave in more human-like ways,” says Miguel Ballesteros, an IBM researcher with the MIT-IBM Watson AI Lab, and co-author of both studies. “The sequential models don’t seem to care if you finish a sentence with a non-grammatical phrase. Why? Because they don’t see that hierarchy.”
As a postdoc at Carnegie Mellon University, Ballesteros helped develop a method for training modern language models on sentence structure called recurrent neural network grammars, or RNNGs. In the current research, he and his colleagues exposed the RNNG model, and similar models with little-to-no grammar training, to sentences with good, bad, or ambiguous syntax. When human subjects are asked to read sentences that sound grammatically off, their surprise is registered by longer response times. For computers, surprise is expressed in probabilities; when low-probability words appear in the place of high-probability words, researchers give the models a higher surprisal score.
They found that the best-performing model — the grammar-enriched RNNG model — showed greater surprisal when exposed to grammatical anomalies; for example, when the word “that” improperly appears instead of “what” to introduce an embedded clause; “I know what the lion devoured at sunrise” is a perfectly natural sentence, but “I know that the lion devoured at sunrise” sounds like it has something missing — because it does.
Linguists call this type of construction a dependency between a filler (a word like who or what) and a gap (the absence of a phrase where one is typically required). Even when more complicated constructions of this type are shown to grammar-enriched models, they — like native speakers of English — clearly know which ones are wrong.
For example, “The policeman who the criminal shot the politician with his gun shocked during the trial” is anomalous; the gap corresponding to the filler “who” should come after the verb, “shot,” not “shocked.” Rewriting the sentence to change the position of the gap, as in “The policeman who the criminal shot with his gun shocked the jury during the trial,” is longwinded, but perfectly grammatical.
“Without being trained on tens of millions of words, state-of-the-art sequential models don’t care where the gaps are and aren’t in sentences like those,” says Roger Levy, a professor in MIT’s Department of Brain and Cognitive Sciences, and co-author of both studies. “A human would find that really weird, and, apparently, so do grammar-enriched models.”
Bad grammar, of course, not only sounds weird, it can turn an entire sentence into gibberish, underscoring the importance of syntax in cognition, and to psycholinguists who study syntax to learn more about the brain’s capacity for symbolic thought.“Getting the structure right is important to understanding the meaning of the sentence and how to interpret it,” says Peng Qian, a graduate student at MIT and co-author of both studies.
The researchers plan to next run their experiments on larger datasets and find out if grammar-enriched models learn new words and phrases faster. Just as submitting neural networks to psychology tests is helping AI engineers understand and improve language models, psychologists hope to use this information to build better models of the brain.
“Some component of our genetic endowment gives us this rich ability to speak,” says Ethan Wilcox, a graduate student at Harvard and co-author of both studies. “These are the sorts of methods that can produce insights into how we learn and understand language when our closest kin cannot.”
Wearing a sensor-packed glove while handling a variety of objects, MIT researchers have compiled a massive dataset that enables an AI system to recognize objects through touch alone. The information could be leveraged to help robots identify and manipulate objects, and may aid in prosthetics design.
The researchers developed a low-cost knitted glove, called “scalable tactile glove” (STAG), equipped with about 550 tiny sensors across nearly the entire hand. Each sensor captures pressure signals as humans interact with objects in various ways. A neural network processes the signals to “learn” a dataset of pressure-signal patterns related to specific objects. Then, the system uses that dataset to classify the objects and predict their weights by feel alone, with no visual input needed.
In a paper published today in Nature, the researchers describe a dataset they compiled using STAG for 26 common objects — including a soda can, scissors, tennis ball, spoon, pen, and mug. Using the dataset, the system predicted the objects’ identities with up to 76 percent accuracy. The system can also predict the correct weights of most objects within about 60 grams.
Similar sensor-based gloves used today run thousands of dollars and often contain only around 50 sensors that capture less information. Even though STAG produces very high-resolution data, it’s made from commercially available materials totaling around $10.
The tactile sensing system could be used in combination with traditional computer vision and image-based datasets to give robots a more human-like understanding of interacting with objects.
“Humans can identify and handle objects well because we have tactile feedback. As we touch objects, we feel around and realize what they are. Robots don’t have that rich feedback,” says Subramanian Sundaram PhD ’18, a former graduate student in the Computer Science and Artificial Intelligence Laboratory (CSAIL). “We’ve always wanted robots to do what humans can do, like doing the dishes or other chores. If you want robots to do these things, they must be able to manipulate objects really well.”
The researchers also used the dataset to measure the cooperation between regions of the hand during object interactions. For example, when someone uses the middle joint of their index finger, they rarely use their thumb. But the tips of the index and middle fingers always correspond to thumb usage. “We quantifiably show, for the first time, that, if I’m using one part of my hand, how likely I am to use another part of my hand,” he says.
Prosthetics manufacturers can potentially use information to, say, choose optimal spots for placing pressure sensors and help customize prosthetics to the tasks and objects people regularly interact with.
Joining Sundaram on the paper are: CSAIL postdocs Petr Kellnhofer and Jun-Yan Zhu; CSAIL graduate student Yunzhu Li; Antonio Torralba, a professor in EECS and director of the MIT-IBM Watson AI Lab; and Wojciech Matusik, an associate professor in electrical engineering and computer science and head of the Computational Fabrication group.
STAG is laminated with an electrically conductive polymer that changes resistance to applied pressure. The researchers sewed conductive threads through holes in the conductive polymer film, from fingertips to the base of the palm. The threads overlap in a way that turns them into pressure sensors. When someone wearing the glove feels, lifts, holds, and drops an object, the sensors record the pressure at each point.
The threads connect from the glove to an external circuit that translates the pressure data into “tactile maps,” which are essentially brief videos of dots growing and shrinking across a graphic of a hand. The dots represent the location of pressure points, and their size represents the force — the bigger the dot, the greater the pressure.
From those maps, the researchers compiled a dataset of about 135,000 video frames from interactions with 26 objects. Those frames can be used by a neural network to predict the identity and weight of objects, and provide insights about the human grasp.
To identify objects, the researchers designed a convolutional neural network (CNN), which is usually used to classify images, to associate specific pressure patterns with specific objects. But the trick was choosing frames from different types of grasps to get a full picture of the object.
The idea was to mimic the way humans can hold an object in a few different ways in order to recognize it, without using their eyesight. Similarly, the researchers’ CNN chooses up to eight semirandom frames from the video that represent the most dissimilar grasps — say, holding a mug from the bottom, top, and handle.
But the CNN can’t just choose random frames from the thousands in each video, or it probably won’t choose distinct grips. Instead, it groups similar frames together, resulting in distinct clusters corresponding to unique grasps. Then, it pulls one frame from each of those clusters, ensuring it has a representative sample. Then the CNN uses the contact patterns it learned in training to predict an object classification from the chosen frames.
“We want to maximize the variation between the frames to give the best possible input to our network,” Kellnhofer says. “All frames inside a single cluster should have a similar signature that represent the similar ways of grasping the object. Sampling from multiple clusters simulates a human interactively trying to find different grasps while exploring an object.”
For weight estimation, the researchers built a separate dataset of around 11,600 frames from tactile maps of objects being picked up by finger and thumb, held, and dropped. Notably, the CNN wasn’t trained on any frames it was tested on, meaning it couldn’t learn to just associate weight with an object. In testing, a single frame was inputted into the CNN. Essentially, the CNN picks out the pressure around the hand caused by the object’s weight, and ignores pressure caused by other factors, such as hand positioning to prevent the object from slipping. Then it calculates the weight based on the appropriate pressures.
The system could be combined with the sensors already on robot joints that measure torque and force to help them better predict object weight. “Joints are important for predicting weight, but there are also important components of weight from fingertips and the palm that we capture,” Sundaram says.