For years, the researchers have been developing an interactive data-science system called Northstar, which runs in the cloud but has an interface that supports any touchscreen device, including smartphones and large interactive whiteboards. Users feed the system datasets, and manipulate, combine, and extract features on a user-friendly interface, using their fingers or a digital pen, to uncover trends and patterns.

In a paper being presented at the Association for Computing Machinery (ACM) Special Interest Group on Management of Data (SIGMOD) conference, the researchers — including Tim Kraska, an EECS associate professor and long-time Northstar project lead — detail a new component of Northstar. Called VDS, for “virtual data scientist,” the component instantly generates machine-learning models to run prediction tasks on their datasets. Doctors, for instance, can use the system to help predict which patients are more likely to have certain diseases, while business owners might want to forecast sales. If using an interactive whiteboard, everyone can also collaborate in real-time.  

The aim is to democratize data science by making it easy to do complex analytics, quickly and accurately.

“Even a coffee shop owner who doesn’t know data science should be able to predict their sales over the next few weeks to figure out how much coffee to buy,” says Kraska, the paper's co-author, who is a principal investigator in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and founding co-director of the new Data System and AI Lab (DSAIL). “In companies that have data scientists, there’s a lot of back and forth between data scientists and nonexperts, so we can also bring them into one room to do analytics together.”

VDS is based on an increasingly popular technique in artificial intelligence called automated machine-learning (AutoML), which lets people with limited data-science know-how train AI models to make predictions based on their datasets. Currently, the tool leads the DARPA D3M Automatic Machine Learning competition, which every six months decides on the best-performing AutoML tool.    

Joining Kraska on the paper are: first author Zeyuan Shang, a EECS graduate student, and Emanuel Zgraggen, a postdoc and main contributor of Northstar, both of EECS, CSAIL, and DSAIL; Benedetto Buratti, Yeounoh Chung, Philipp Eichmann, and Eli Upfal, all of Brown; and Carsten Binnig who recently moved from Brown to the Technical University of Darmstadt in Germany.

An “unbounded canvas” for analytics

The new work builds on years of collaboration on Northstar between researchers at MIT and Brown. Over four years, the researchers have published numerous papers detailing components of Northstar, including the interactive interface, operations on multiple platforms, accelerating results, and studies on user behavior.

Northstar starts as a blank, white interface. Users upload datasets into the system, which appear in a “datasets” box on the left. Any data labels will automatically populate a separate “attributes” box below. There’s also an “operators” box that contains various algorithms, as well as the new AutoML tool. All data are stored and analyzed in the cloud.

The researchers like to demonstrate the system on a public dataset that contains information on intensive care unit patients. Consider medical researchers who want to examine co-occurrences of certain diseases in certain age groups. They drag and drop into the middle of the interface a pattern-checking algorithm, which at first appears as a blank box. As input, they move into the box disease features labeled, say, “blood,” “infectious,” and “metabolic.” Percentages of those diseases in the dataset appear in the box. Then, they drag the “age” feature into the interface, which displays a bar chart of the patient’s age distribution. Drawing a line between the two boxes links them together. By circling age ranges, the algorithm immediately computes the co-occurrence of the three diseases among the age range.  

“It’s like a big, unbounded canvas where you can lay out how you want everything,” says Zgraggen, who is the key inventor of Northstar’s interactive interface. “Then, you can link things together to create more complex questions about your data.”

Approximating AutoML

With VDS, users can now also run predictive analytics on that data by getting models custom-fit to their tasks, such as data prediction, image classification, or analyzing complex graph structures.

Using the above example, say the medical researchers want to predict which patients may have blood disease based on all features in the dataset. They drag and drop “AutoML” from the list of algorithms. It’ll first produce a blank box, but with a “target” tab, under which they’d drop the “blood” feature. The system will automatically find best-performing machine-learning pipelines, presented as tabs with constantly updated accuracy percentages. Users can stop the process at any time, refine the search, and examine each model’s errors rates, structure, computations, and other things.

According to the researchers, VDS is the fastest interactive AutoML tool to date, thanks, in part, to their custom “estimation engine.” The engine sits between the interface and the cloud storage. The engine leverages automatically creates several representative samples of a dataset that can be progressively processed to produce high-quality results in seconds.

“Together with my co-authors I spent two years designing VDS to mimic how a data scientist thinks,” Shang says, meaning it instantly identifies which models and preprocessing steps it should or shouldn’t run on certain tasks, based on various encoded rules. It first chooses from a large list of those possible machine-learning pipelines and runs simulations on the sample set. In doing so, it remembers results and refines its selection. After delivering fast approximated results, the system refines the results in the back end. But the final numbers are usually very close to the first approximation.

“For using a predictor, you don’t want to wait four hours to get your first results back. You want to already see what’s going on and, if you detect a mistake, you can immediately correct it. That’s normally not possible in any other system,” Kraska says. The researchers’ previous user study, in fact, “show that the moment you delay giving users results, they start to lose engagement with the system.”

The researchers evaluated the tool on 300 real-world datasets. Compared to other state-of-the-art AutoML systems, VDS’ approximations were as accurate, but were generated within seconds, which is much faster than other tools, which operate in minutes to hours

Next, the researchers are looking to add a feature that alerts users to potential data bias or errors. For instance, to protect patient privacy, sometimes researchers will label medical datasets with patients aged 0 (if they do not know the age) and 200 (if a patient is over 95 years old). But novices may not recognize such errors, which could completely throw off their analytics.  

“If you’re a new user, you may get results and think they’re great,” Kraska says. “But we can warn people that there, in fact, may be some outliers in the dataset that may indicate a problem.”

For additional information on this story, including video clips, please visit the MIT News website.

EECS Professor Erik Demaine

MIT Open Learning

Editor's note: For a related video of Erik DeMaine's class, please visit the MIT News website.

EECS faculty member Erik Demaine is among seven MIT educators honored recently for significant innovations in digital learning.

Demaine, also a principal investigator in the Computer Science and Artificial Intelligence Lab (CSAIL), received a Teaching with Digital Technology Award for his course 6.892 (Fun with Hardness Proofs). Co-sponsored by MIT Open Learning and the Office of the Vice Chancellor, the student-nominated awards are presented to faculty and instructors who have improved teaching and learning at MIT with digital technology.

Demaine's course flipped the traditional classroom model. Instead of lecturing in person, all lectures were posted online and problems were done in class. That approach allowed the students to spend class time working together on collaborative problem solving through an online application that Demaine created, called Coauthor.

MIT students nominated 117 faculty and instructors for this award this year, more than in any previous year. Other winners included John Belcher, Class of '22 Professor of Physics;  Amy Carleton, lecturer in Comparative Media Studies/Writing; and Jared Curhan, associate professor of work and organization studies at the MIT Sloan School of Management.

Three other MIT educators shared the third annual MITx Prize for Teaching and Learning in MOOCs for their work in developing massive open opline courses (better known as MOOCs). They are: Polina Anikeeva, associate professor in the Department of Materials Science and Engineering (DMSE); Martin Z. Bazant, E.G. Roos (1944) Professor of Chemical Engineering; and Jessica Sandland, a DMSE lecturer and MITx Digital Learning Scientist.

For an extended report on all the digital-learning award winners, please visit the MIT News website.

Fernando “Corby” Corbató, an MIT professor emeritus whose work in the 1960s on time-sharing systems broke important ground in democratizing the use of computers, died on Friday, July 12, at his home in Newburyport, Massachusetts. He was 93.

Decades before the existence of concepts like cybersecurity and the cloud, Corbató led the development of one of the world’s first operating systems. His “Compatible Time-Sharing System” (CTSS) allowed multiple people to use a computer at the same time, greatly increasing the speed at which programmers could work. It’s also widely credited as the first computer system to use passwords. 

After CTSS Corbató led a time-sharing effort called Multics, which directly inspired operating systems like Linux and laid the foundation for many aspects of modern computing. Multics doubled as a fertile training ground for an emerging generation of programmers that included C programming language creator Dennis Ritchie, Unix developer Ken Thompson, and spreadsheet inventors Dan Bricklin and Bob Frankston.

Before time-sharing, using a computer was tedious and required detailed knowledge. Users would create programs on cards and submit them in batches to an operator, who would enter them to be run one at a time over a series of hours. Minor errors would require repeating this sequence, often more than once.

But with CTSS, which was first demonstrated in 1961, answers came back in mere seconds, forever changing the model of program development. Decades before the PC revolution, Corbató and his colleagues also opened up communication between users with early versions of email, instant messaging, and word processing. 

“Corby was one of the most important researchers for making computing available to many people for many purposes,” says long-time colleague Tom Van Vleck. “He saw that these concepts don’t just make things more efficient; they fundamentally change the way people use information.”

Besides making computing more efficient, CTSS also inadvertently helped establish the very concept of digital privacy itself. With different users wanting to keep their own files private, CTSS introduced the idea of having people create individual accounts with personal passwords. Corbató’s vision of making high-performance computers available to more people also foreshadowed trends in cloud computing, in which tech giants like Amazon and Microsoft rent out shared servers to companies around the world. 

“Other people had proposed the idea of time-sharing before,” says Jerry Saltzer, who worked on CTSS with Corbató after starting out as his teaching assistant. “But what he brought to the table was the vision and the persistence to get it done.”

CTSS was also the spark that convinced MIT to launch “Project MAC,” the precursor to the Laboratory for Computer Science (LCS). LCS later merged with the Artificial Intelligence Lab to become MIT’s largest research lab, the Computer Science and Artificial Intelligence Laboratory (CSAIL), which is now home to more than 600 researchers. 

“It’s no overstatement to say that Corby’s work on time-sharing fundamentally transformed computers as we know them today,” says CSAIL Director Daniela Rus. “From PCs to smartphones, the digital revolution can directly trace its roots back to the work that he led at MIT nearly 60 years ago.” 

In 1990, Corbató was honored for his work with the Association of Computing Machinery’s Turing Award, often described as “the Nobel Prize for computing.”

From sonar to CTSS

Corbató was born on July 1, 1926, in Oakland, California. At 17, he enlisted as a technician in the U.S. Navy, where he first got the engineering bug working on a range of radar and sonar systems. After World War II he earned his bachelor's degree at Caltech before heading to MIT to complete a PhD in physics. 

As a PhD student, Corbató met Professor Philip Morse, who recruited him to work with his team on Project Whirlwind, the first computer capable of real-time computation. After graduating, Corbató joined MIT's Computation Center as a research assistant, soon moving up to become deputy director of the entire center. 

It was there that he started thinking about ways to make computing more efficient. For all its innovation, Whirlwind was still a rather clunky machine. Researchers often had trouble getting much work done on it, since they had to take turns using it for half-hour chunks of time. (Corbató said that it had a habit of crashing every 20 minutes or so.) 

Since computer input and output devices were much slower than the computer itself, in the late 1950s a scheme called multiprogramming was developed to allow a second program to run whenever the first program was waiting for some device to finish. Time-sharing built on this idea, allowing other programs to run while the first program was waiting for a human user to type a request, thus allowing the user to interact directly with the first program.

Saltzer says that Corbató pioneered a programming approach that would be described today as agile design. 

“It’s a buzzword now, but back then it was just this iterative approach to coding that Corby encouraged and that seemed to work especially well,” he says.  

In 1962, Corbató published a paper about CTSS that quickly became the talk of the slowly-growing computer science community. The following year, MIT invited several hundred programmers to campus to try out the system, spurring a flurry of further research on time-sharing.

Foreshadowing future technological innovation, Corbató was amazed — and amused — by how quickly people got habituated to CTSS’ efficiency.

“Once a user gets accustomed to [immediate] computer response, delays of even a fraction of a minute are exasperatingly long,” he presciently wrote in his 1962 paper. “First indications are that programmers would readily use such a system if it were generally available.”

Multics, meanwhile, expanded on CTSS’ more ad hoc design with a hierarchical file system, better interfaces to email and instant messaging, and more precise privacy controls. Peter Neumann, who worked at Bell Labs when they were collaborating with MIT on Multics, says that its design prevented the possibility of many vulnerabilities that impact modern systems, like “buffer overflow” (which happens when a program tries to write data outside the computer’s short-term memory). 

“Multics was so far ahead of the rest of the industry,” says Neumann. “It was intensely software-engineered, years before software engineering was even viewed as a discipline.” 

In spearheading these time-sharing efforts, Corbató served as a soft-spoken but driven commander in chief — a logical thinker who led by example and had a distinctly systems-oriented view of the world.

“One thing I liked about working for Corby was that I knew he could do my job if he wanted to,” says Van Vleck. “His understanding of all the gory details of our work inspired intense devotion to Multics, all while still being a true gentleman to everyone on the team.” 

Another legacy of the professor’s is “Corbató’s Law,” which states that the number of lines of code someone can write in a day is the same regardless of the language used. This maxim is often cited by programmers when arguing in favor of using higher-level languages.

Corbató was an active member of the MIT community, serving as associate department head for computer science and engineering from 1974 to 1978 and 1983 to 1993. He was a member of the National Academy of Engineering, and a fellow of the Institute of Electrical and Electronics Engineers and the American Association for the Advancement of Science. 

Corbató is survived by his wife, Emily Corbató, from Brooklyn, New York; his stepsons, David and Jason Gish; his brother, Charles; and his daughters, Carolyn and Nancy, from his marriage to his late wife Isabel; and five grandchildren. 

In lieu of flowers, gifts may be made to MIT’s Fernando Corbató Fellowship Fund via Bonny Kellermann in the Memorial Gifts Office. 

CSAIL will host an event to honor and celebrate Corbató in the coming months.

For additional information on Professor Corbató's life and legacy, please visit the MIT News website.
 

Patrick Winston, a beloved professor and computer scientist at MIT, died on July 19 at Massachusetts General Hospital in Boston. He was 76.
 
A professor at MIT for almost 50 years, Winston was director of MIT’s Artificial Intelligence Laboratory from 1972 to 1997 before it merged with the Laboratory for Computer Science to become MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL).
 
A devoted teacher and cherished colleague, Winston led CSAIL’s Genesis Group, which focused on developing AI systems that have human-like intelligence, including the ability to tell, perceive, and comprehend stories. He believed that such work could help illuminate aspects of human intelligence that scientists don’t yet understand.
 
“My principal interest is in figuring out what’s going on inside our heads, and I’m convinced that one of the defining features of human intelligence is that we can understand stories,'” Winston, the Ford Professor of Artificial Intelligence and Computer Science, said in a 2011 interview for CSAIL. “Believing as I do that stories are important, it was natural for me to try to build systems that understand stories, and that shed light on what the story-understanding process is all about.”
 
He was renowned for his accessible and informative lectures, and gave a hugely popular talk every year during the Independent Activities Period called “How to Speak.” 
 
“As a speaker he always had his audience in the palm of his hand,” says MIT Professor Peter Szolovits. “He put a tremendous amount of work into his lectures, and yet managed to make them feel loose and spontaneous. He wasn’t flashy, but he was compelling and direct. ”
 
Winston’s dedication to teaching earned him many accolades over the years, including the Baker Award, the Eta Kappa Nu Teaching Award, and the Graduate Student Council Teaching Award.
 
“Patrick’s humanity and his commitment to the highest principles made him the soul of EECS,” MIT President L. Rafael Reif wrote in a letter to the MIT community. “I called on him often for advice and feedback, and he always responded with kindness, candor, wisdom and integrity.  I will be forever grateful for his counsel, his objectivity, and his tremendous inspiration and dedication to our students.”
 
Teaching computers to think

Born Feb. 5, 1943, in Peoria, Illinois, Winston was always exceptionally curious about science, technology and how to use such tools to explore what it means to be human. He was an MIT "lifer" starting in 1961, earning his bachelor’s, master’s and doctoral degrees from the Institute before joining the faculty of the Department of Electrical Engineering and Computer Science in 1970.
 
His thesis work with Marvin Minsky centered on the difficulty of learning, setting off a trajectory of work where he put a playful, yet laser-sharp focus on fine-tuning AI systems to better understand stories.
 
His Genesis project aimed to faithfully model computers after human intelligence in order to fully grasp the inner workings of our own motivations, rationality, and perception. Using MIT research scientist Boris Katz’s START natural language processing system and a vision system developed by former MIT PhD student Sajit Rao, Genesis can digest short, simple chunks of text, then spit out reports about how it interpreted connections between events.
 
While the system has processed many works, Winston chose “Macbeth” as a primary text because the tragedy offers an opportunity to take big human themes, such as greed and revenge, and map out their components.
 
“[Shakespeare] was pretty good at his portrayal of ‘the human condition,’ as my friends in the humanities would say,” Winston told The Boston Globe. “So there’s all kinds of stuff in there about what’s typical when we humans wander through the world.”
 
His deep fascination with humanity, human intelligence, and how we communicate information spilled over into what he often described as his favorite academic activity: teaching.
 
“He was a superb educator who introduced the field to generations of students,” says EECS Professor and longtime colleague Randall Davis. “His lectures had an uncanny ability to move in minutes from the details of an algorithm to the larger issues it illustrated, to yet larger lessons about how to be a scientist and a human being.”
 
A past president of the Association for the Advancement of Artificial Intelligence (AAAI), Winston also wrote and edited numerous books, including a seminal textbook on AI that’s still used in classrooms around the world. Outside of the lab he also co-founded Ascent Technology, which produces scheduling and workforce management applications for major airports.
 
He is survived by his wife Karen Prendergast and his daughter Sarah.

MIT professors David Sontag and Peter Szolovits don’t assign a textbook for their class, 6.S897/HST.956 (Machine Learning for Healthcare), because there isn’t one. Instead, students read scientific papers, solve problem sets based on current topics like opioid addiction and infant mortality, and meet the doctors and engineers paving the way for a more data-driven approach to health care. Jointly offered by MIT’s Department of Electrical Engineering and Computer Science (EECS) and the Harvard-MIT program in Health Sciences Technology, the class is one of just a handful offered across the country.

“Because it’s a new field, what we teach will help shape how AI is used to diagnose and treat patients,” says Irene Chen, an EECS graduate student who helped design and teach the course. “We tried to give students the freedom to be creative and explore the many ways machine learning is being applied to health care.”

Two-thirds of the syllabus this spring was new. Students were introduced to the latest machine-learning algorithms for analyzing doctors’ clinical notes, patient medical scans, and electronic health records, among other data. Students also explored the risks of using automated methods to explore large, often messy observational datasets, from confusing correlation with causation to understanding how AI models can make bad decisions based on biased data or faulty assumptions.

With all of the hype around AI, the course had more takers than seats. After 100 students showed up on the first day, students were assigned a quiz to test their knowledge of statistics and other prerequisites. That helped whittle the class down to 70. Michiel Bakker, a graduate student at the MIT Media Lab, made the cut and says the course gave him medical concepts that most engineering courses don’t provide.

“In machine learning, the data are often either images or text,” he says. “Here we learned the importance of combining genetic data with medical images with electronic health records. To use machine learning in health care you have to understand the problems, how to combine techniques and anticipate where things could go wrong.”

Most lectures and homework problems focused on real world scenarios, drawing from MIT’s MIMIC critical care database and a subset of the IBM MarketScan Research Databases focused on insurance claims. The course also featured regular guest lectures by Boston-area clinicians. In a reversal of roles, students held office hours for doctors interested in integrating AI into their practice. 

“There are so many people in academia working on machine learning, and so many doctors at hospitals in Boston,” says Willie Boag, an EECS graduate student who helped design and teach the course. “There’s so much opportunity in fostering conversation between these groups.”

In health care, as in other fields where AI has made inroads, regulators are discussing what rules should be put in place to protect the public. The U.S. Federal Drug Administration recently released a draft framework for regulating AI products, which students got to review and comment on, in class and in feedback published online in the Federal Register.

Andy Coravos, a former entrepreneur in residence at the FDA, now CEO of Elektra Labs in Boston, helped lead the discussion and was impressed by the quality of the comments. “Many students identified test cases relevant to the current white paper, and used those examples to draft public comments for what to keep, add, and change in future iterations,” she says.

The course culminated in a final project in which teams of students used the MIMIC and IBM datasets to explore a timely question in the field. One team analyzed insurance claims to explore regional variation in screening patients for early-stage kidney disease. Many patients with hypertension and diabetes are never tested for chronic kidney disease, even though both conditions put them at high risk. The students found that they could predict fairly well who would be screened, and that screening rates diverged most between the southern and northeastern United States.

“If this work were to continue, the next step would be to share the results with a doctor and get their perspective,” says team member Matt Groh, a PhD student at the MIT Media Lab. “You need that cross-disciplinary feedback.”

The MIT-IBM Watson AI Lab made the anonymized data available, and provided student-access to the IBM cloud, out of an interest in helping to educate the next generation of scientists and engineers, says Kush Varshney, principal research staff member and manager at IBM Research. “Health care is messy and complex, which is why there are no substitutes for working with real-world data,” he says.

Szolovits agrees. Using synthetic data would have been easier but far less meaningful. “It’s important for students to grapple with the complexities of real data,” he says. “Any researcher developing automated techniques and tools to improve patient care needs to be sensitive to its many nuances.” (In May 2019, Szolovits and Sontag received the Burgess (1952) and Elizabeth Jamieson Prize for Excellence in Teaching in recognition of their work in developing the class.)

In a recent recap on Twitter, Chen gave shout-outs to the students, guest lecturers, professors, and her fellow teaching assistant. She also reflected on the joys of teaching. “Research is rewarding and often fun, but helping someone see your field with fresh eyes is insanely cool.”

A novel sensor designed by MIT researchers could dramatically accelerate the process of diagnosing sepsis, a leading cause of death in U.S. hospitals that kills nearly 250,000 patients annually.

Sepsis occurs when the body’s immune response to infection triggers an inflammation chain reaction throughout the body, causing high heart rate, high fever, shortness of breath, and other issues. If left unchecked, it can lead to septic shock, where blood pressure falls and organs shut down. To diagnose sepsis, doctors traditionally rely on various diagnostic tools, including vital signs, blood tests, and other imaging and lab tests.

In recent years, researchers have found protein biomarkers in the blood that are early indicators of sepsis. One promising candidate is interleukin-6 (IL-6), a protein produced in response to inflammation. In sepsis patients, IL-6 levels can rise hours before other symptoms begin to show. But even at these elevated levels, the concentration of this protein in the blood is too low overall for traditional assay devices to detect it quickly.

In a paper being presented this week at the Engineering in Medicine and Biology Conference, MIT researchers describe a microfluidics-based system that automatically detects clinically significant levels of IL-6 for sepsis diagnosis in about 25 minutes, using less than a finger prick of blood.

In one microfluidic channel, microbeads laced with antibodies mix with a blood sample to capture the IL-6 biomarker. In another channel, only beads containing the biomarker attach to an electrode. Running voltage through the electrode produces an electrical signal for each biomarker-laced bead, which is then converted into the biomarker concentration level.

“For an acute disease, such as sepsis, which progresses very rapidly and can be life-threatening, it’s helpful to have a system that rapidly measures these nonabundant biomarkers,” says first author Dan Wu, a PhD student in the Department of Mechanical Engineering. “You can also frequently monitor the disease as it progresses.”

Joining Wu on the paper is Joel Voldman, a professor and associate head of the Department of Electrical Engineering and Computer Science, co-director of the Medical Electronic Device Realization Center, and a principal investigator in the Research Laboratory of Electronics and the Microsystems Technology Laboratories.

Integrated, automated design

Traditional assays that detect protein biomarkers are bulky, expensive machines relegated to labs that require about a milliliter of blood and produce results in hours. In recent years, portable “point-of-care” systems have been developed that use microliters of blood to get similar results in about 30 minutes.

But point-of-care systems can be very expensive since most use pricey optical components to detect the biomarkers. They also capture only a small number of proteins, many of which are among the more abundant ones in blood. Any efforts to decrease the price, shrink down components, or increase protein ranges negatively impacts their sensitivity.

In their work, the researchers wanted to shrink components of the magnetic-bead-based assay, which is often used in labs, onto an automated microfluidics device that’s roughly several square centimeters. That required manipulating beads in micron-sized channels and fabricating a device in the Microsystems Technology Laboratory that automated the movement of fluids.

The beads are coated with an antibody that attracts IL-6, as well as a catalyzing enzyme called horseradish peroxidase. The beads and blood sample are injected into the device, entering into an “analyte-capture zone,” which is basically a loop. Along the loop is a peristaltic pump — commonly used for controlling liquids — with valves automatically controlled by an external circuit. Opening and closing the valves in specific sequences circulates the blood and beads to mix together. After about 10 minutes, the IL-6 proteins have bound to the antibodies on the beads.

Automatically reconfiguring the valves at that time forces the mixture into a smaller loop, called the “detection zone,” where they stay trapped. A tiny magnet collects the beads for a brief wash before releasing them around the loop. After about 10 minutes, many beads have stuck on an electrode coated with a separate antibody that attracts IL-6. At that time, a solution flows into the loop and washes the untethered beads, while the ones with IL-6 protein remain on the electrode.

The solution carries a specific molecule that reacts to the horseradish enzyme to create a compound that responds to electricity. When a voltage is applied to the solution, each remaining bead creates a small current. A common chemistry technique called “amperometry” converts that current into a readable signal. The device counts the signals and calculates the concentration of IL-6.

“On their end, doctors just load in a blood sample using a pipette. Then, they press a button and 25 minutes later they know the IL-6 concentration,” Wu says.

The device uses about 5 microliters of blood, which is about a quarter the volume of blood drawn from a finger prick and a fraction of the 100 microliters required to detect protein biomarkers in lab-based assays. The device captures IL-6 concentrations as low as 16 picograms per milliliter, which is below the concentrations that signal sepsis, meaning the device is sensitive enough to provide clinically relevant detection.

A general platform

The current design has eight separate microfluidics channels to measure as many different biomarkers or blood samples in parallel. Different antibodies and enzymes can be used in separate channels to detect different biomarkers, or different antibodies can be used in the same channel to detect several biomarkers simultaneously.

Next, the researchers plan to create a panel of important sepsis biomarkers for the device to capture, including interleukin-6, interleukin-8, C-reactive protein, and procalcitonin. But there’s really no limit to how many different biomarkers the device can measure, for any disease, Wu says. Notably, more than 200 protein biomarkers for various diseases and conditions have been approved by the U.S. Food and Drug Administration.

“This is a very general platform,” Wu says. “If you want to increase the device’s physical footprint, you can scale up and design more channels to detect as many biomarkers as you want.”

The work was funded by Analog Devices, Maxim Integrated, and the Novartis Institutes of Biomedical Research.

Behind the scenes of every web service, from a secure web browser to an entertaining app, is a programmer’s code, carefully written to ensure everything runs quickly, smoothly, and securely. For years, EECS Associate Professor Adam Chlipala has been toiling away behind behind-the-scenes, developing tools to help programmers more quickly and easily generate their code — and prove it does what it’s supposed to do.

Scanning the many publications on Chlipala’s webpage, you’ll find some commonly repeated keywords, such as “easy,” “automated,” and “proof.” Much of his work centers on designing simplified programming languages and app-making tools for programmers, systems that automatically generate optimized algorithms for specific tasks, and compilers that automatically prove that the complex math written in code is correct.

“I hope to save a lot of people a lot of time doing boring repetitive work, by automating programming work as well as decreasing the cost of building secure, reliable systems,” says Chlipala, who is a recently tenured professor of computer science, a principal investigator in the Computer Science and Artificial Intelligence Laboratory (CSAIL), and head of the Programming Languages and Verification Group.

One of Chlipala’s recent systems automatically generates optimized — and mathematically proven — cryptographic algorithms, freeing programmers from hours upon hours of manually writing and verifying code by hand. And that system is now behind nearly all secure Google Chrome communications.

But Chlipala’s code-generating and mathematical proof systems can be used for a wide range of applications, from protecting financial transactions against fraud to ensuring autonomous vehicles operate safely. The aim, he says, is catching coding errors before they lead to real-world consequences.

“Today, we just assume that there’s going to be a constant flow of serious security problems in all major operating systems. But using formal mathematical methods, we should be able to automatically guarantee there will be far fewer surprises of that kind,” he says. “With a fixed engineering budget, we can suddenly do a lot more, without causing embarrassing or life-threatening disasters.”

A heart for system infrastructure

As he was growing up in the Lehigh Valley region of Pennsylvania, programming became “an important part of my self-identity,” Chlipala says. In the late 1980s, when Chlipala was young, his father, a researcher who ran physics experiments for AT&T Bell Laboratories, taught him some basic programming skills. He quickly became hooked.

In the late 1990s, when the family finally connected to the internet, Chlipala had access to various developer resources that helped him delve “into more serious stuff,” meaning designing larger, more complex programs. He worked on compilers — programs that translate programming language into machine-readable code — and web applications, “when apps were an avant-garde subject.”  

In fact, apps were then called “CGI scripts.” CGI is an acronym for Common Gateway Interface, which is a protocol that enables a program (or “script”) to talk to a server. In high school, Chlipala and some friends designed CGI scripts that connected them in an online forum for young programmers. “It was a means for us to start building our own system infrastructure,” he says.

And as an avid computer gamer, the logical thing for a teenaged Chlipala to do was design his own games. His first attempts were text-based adventures coded in the BASIC programming language. Later, in the C programming language, he designed a “Street Fighter”-like game, called Brimstone, and some simulated combat tabletop games.

It was exciting stuff for a high schooler. “But my heart was always in systems infrastructure, like code compilers and building help tools for old Windows operating systems,” Chlipala says.

From then on, Chlipala worked far in the background of web services, building the programming foundations for developers. “I’m several levels of abstraction removed from the type of computer programming that’s of any interest to any end-user,” he says, laughing.

Impact in the real world

After high school, in 2000, Chlipala enrolled at Carnegie Mellon University, where he majored in computer science and got involved in a programming language compiler research group. In 2007, he earned his PhD in computer science from University of California at Berkeley, where his work focused on developing methods that can prove the mathematical correctness of algorithms.

After completing a postdoc at Harvard University, Chlipala came to MIT in 2011 to begin his teaching career. What drew Chlipala to MIT, in part, was an opportunity “to plug in a gap, where no one was doing my kind of proofs of computer systems’ correctness,” he says. “I enjoyed building that subject here from the ground up.”

Testing the source code that powers web services and computer systems today is computationally intensive. It mostly relies on running the code through tons of simulations, and correcting any caught bugs, until the code produces a desired output. But it’s nearly impossible to run the code through every possible scenario to prove it’s completely without error.

Chlipala’s research group instead focuses on eliminating the need for those simulations, by designing proven mathematical theorems that capture exactly how a given web service or computer system is supposed to behave. From that, they build algorithms that check if the source code operates according to that theorem, meaning it performs exactly how it’s supposed to, mostly during code compiling.

Even though such methods can be applied to any application, Chlipala likes to run his research group like a startup, encouraging students to target specific, practical applications for their research projects. In fact, two of his former students recently joined startups doing work connected to their thesis research.  

One student is working on developing a platform that lets people rapidly design, fabricate, and test their own computer chips. Another is designing mathematical proven systems to ensure the source code powering driverless car systems doesn’t contain errors that’ll lead to mistakes on the road. “In driverless cars, a bug could literally cause a crash, not just the ‘blue-screen death’ type of a crash,” Chlipala says.

Now on sabbatical from this summer until the end of the year, Chlipala is splitting his time between MIT research projects and launching his own startup based around tools that help people without programming experience create advanced apps. One such tool, which lets nonexperts build scheduling apps, has already found users among faculty and staff in his own department. About the new company, he says: “I’ve been into entrepreneurship over the last few years. But now that I have tenure, it’s a good time to get started.”

A novel system developed by MIT researchers automatically “learns” how to schedule data-processing operations across thousands of servers — a task traditionally reserved for imprecise, human-designed algorithms. Doing so could help today’s power-hungry data centers run far more efficiently.

Data centers can contain tens of thousands of servers, which constantly run data-processing tasks from developers and users. Cluster scheduling algorithms allocate the incoming tasks across the servers, in real-time, to efficiently utilize all available computing resources and get jobs done fast.

Traditionally, however, humans fine-tune those scheduling algorithms, based on some basic guidelines (“policies”) and various tradeoffs. They may, for instance, code the algorithm to get certain jobs done quickly or split resource equally between jobs. But workloads — meaning groups of combined tasks — come in all sizes. Therefore, it’s virtually impossible for humans to optimize their scheduling algorithms for specific workloads and, as a result, they often fall short of their true efficiency potential.

The MIT researchers instead offloaded all of the manual coding to machines. In a paper presented at a recent conference of the Association for Computing Machinery (ACM) special interest group on data communiations (SIGCOMM), they describe a system that leverages “reinforcement learning” (RL), a trial-and-error machine-learning technique, to tailor scheduling decisions to specific workloads in specific server clusters.

To do so, they built novel RL techniques that could train on complex workloads. In training, the system tries many possible ways to allocate incoming workloads across the servers, eventually finding an optimal tradeoff in utilizing computation resources and quick processing speeds. No human intervention is required beyond a simple instruction, such as, “minimize job-completion times.”

Compared to the best handwritten scheduling algorithms, the researchers’ system completes jobs about 20 to 30 percent faster, and twice as fast during high-traffic times. Mostly, however, the system learns how to compact workloads efficiently to leave little waste. Results indicate the system could enable data centers to handle the same workload at higher speeds, using fewer resources.

“If you have a way of doing trial and error using machines, they can try different ways of scheduling jobs and automatically figure out which strategy is better than others,” says Hongzi Mao, a PhD student in the Department of Electrical Engineering and Computer Science (EECS). “That can improve the system performance automatically. And any slight improvement in utilization, even 1 percent, can save millions of dollars and a lot of energy in data centers.”

“There’s no one-size-fits-all to making scheduling decisions,” adds co-author Mohammad Alizadeh, an EECS professor and researcher in the Computer Science and Artificial Intelligence Laboratory (CSAIL). “In existing systems, these are hard-coded parameters that you have to decide up front. Our system instead learns to tune its schedule policy characteristics, depending on the data center and workload.”

Joining Mao and Alizadeh on the paper are: postdocs Malte Schwarzkopf and Shaileshh Bojja Venkatakrishnan, and graduate research assistant Zili Meng, all of CSAIL.

RL for scheduling

Typically, data processing jobs come into data centers represented as graphs of “nodes” and “edges.” Each node represents some computation task that needs to be done, where the larger the node, the more computation power needed. The edges connecting the nodes link connected tasks together. Scheduling algorithms assign nodes to servers, based on various policies.

But traditional RL systems are not accustomed to processing such dynamic graphs. These systems use a software “agent” that makes decisions and receives a feedback signal as a reward. Essentially, it tries to maximize its rewards for any given action to learn an ideal behavior in a certain context. They can, for instance, help robots learn to perform a task like picking up an object by interacting with the environment, but that involves processing video or images through an easier set grid of pixels.

To build their RL-based scheduler, called Decima, the researchers had to develop a model that could process graph-structured jobs, and scale to a large number of jobs and servers. Their system’s “agent” is a scheduling algorithm that leverages a graph neural network, commonly used to process graph-structured data. To come up with a graph neural network suitable for scheduling, they implemented a custom component that aggregates information across paths in the graph — such as quickly estimating how much computation is needed to complete a given part of the graph. That’s important for job scheduling, because “child” (lower) nodes cannot begin executing until their “parent” (upper) nodes finish, so anticipating future work along different paths in the graph is central to making good scheduling decisions.

To train their RL system, the researchers simulated many different graph sequences that mimic workloads coming into data centers. The agent then makes decisions about how to allocate each node along the graph to each server. For each decision, a component computes a reward based on how well it did at a specific task — such as minimizing the average time it took to process a single job. The agent keeps going, improving its decisions, until it gets the highest reward possible.

Baselining workloads

One concern, however, is that some workload sequences are more difficult than others to process, because they have larger tasks or more complicated structures. Those will always take longer to process — and, therefore, the reward signal will always be lower — than simpler ones. But that doesn’t necessarily mean the system performed poorly: It could make good time on a challenging workload but still be slower than an easier workload. That variability in difficulty makes it challenging for the model to decide what actions are good or not.

To address that, the researchers adapted a technique called “baselining” in this context. This technique takes averages of scenarios with a large number of variables and uses those averages as a baseline to compare future results. During training, they computed a baseline for every input sequence. Then, they let the scheduler train on each workload sequence multiple times. Next, the system took the average performance across all of the decisions made for the same input workload. That average is the baseline against which the model could then compare its future decisions to determine if its decisions are good or bad. They refer to this new technique as “input-dependent baselining.”

That innovation, the researchers say, is applicable to many different computer systems. “This is general way to do reinforcement learning in environments where there’s this input process that effects environment, and you want every training event to consider one sample of that input process,” he says. “Almost all computer systems deal with environments where things are constantly changing.”

Aditya Akella, a professor of computer science at the University of Wisconsin at Madison, whose group has designed several high-performance schedulers, found the MIT system could help further improve their own policies. “Decima can go a step further and find opportunities for [scheduling] optimization that are simply too onerous to realize via manual design/tuning processes,” Akella says. “The schedulers we designed achieved significant improvements over techniques used in production in terms of application performance and cluster efficiency, but there was still a gap with the ideal improvements we could possibly achieve. Decima shows that an RL-based approach can discover [policies] that help bridge the gap further. Decima improved on our techniques by a [roughly] 30 percent, which came as a huge surprise.”

Right now, their model is trained on simulations that try to recreate incoming online traffic in real-time. Next, the researchers hope to train the model on real-time traffic, which could potentially crash the servers. So, they’re currently developing a “safety net” that will stop their system when it’s about to cause a crash. “We think of it as training wheels,” Alizadeh says. “We want this system to continuously train, but it has certain training wheels that if it goes too far we can ensure it doesn’t fall over.”

The field of predictive analytics holds increasing promise for helping clinicians diagnose and treat patients. Machine-learning models can be trained to find patterns in patient data to aid in sepsis care, design safer chemotherapy regimens, and predict a patient’s risk of having breast cancer or dying in the ICU, to name just a few examples.

Typically, training datasets consist of many sick and healthy subjects, but with relatively little data for each subject. Experts must then find just those aspects — or “features” — in the datasets that will be important for making predictions.

This “feature engineering” can be a laborious and expensive process. But it’s becoming even more challenging with the rise of wearable sensors, because researchers can more easily monitor patients’ biometrics over long periods, tracking sleeping patterns, gait, and voice activity, for example. After only a week’s worth of monitoring, experts could have several billion data samples for each subject.  

In a paper presented at the Machine Learning for Healthcare conference recently, MIT researchers demonstrate a model that automatically learns features predictive of vocal cord disorders. The features come from a dataset of about 100 subjects, each with about a week’s worth of voice-monitoring data and several billion samples — in other words, a small number of subjects and a large amount of data per subject. The dataset contain signals captured from a little accelerometer sensor mounted on subjects’ necks.

In experiments, the model used features automatically extracted from these data to classify, with high accuracy, patients with and without vocal cord nodules. These are lesions that develop in the larynx, often because of patterns of voice misuse such as belting out songs or yelling. Importantly, the model accomplished this task without a large set of hand-labeled data.

“It’s becoming increasing easy to collect long time-series datasets. But you have physicians that need to apply their knowledge to labeling the dataset,” says lead author Jose Javier Gonzalez Ortiz, a PhD student in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). “We want to remove that manual part for the experts and offload all feature engineering to a machine-learning model.”

The model can be adapted to learn patterns of any disease or condition. But the ability to detect the daily voice-usage patterns associated with vocal cord nodules is an important step in developing improved methods to prevent, diagnose, and treat the disorder, the researchers say. That could include designing new ways to identify and alert people to potentially damaging vocal behaviors.

Joining Gonzalez Ortiz on the paper are JohnGuttag, the Dugald C. Jackson Professor of Computer Science and Electrical Engineering and head of CSAIL’s Data Driven Inference Group; Robert Hillman, Jarrad Van Stan, and Daryush Mehta, all of Massachusetts General Hospital’s Center for Laryngeal Surgery and Voice Rehabilitation; and Marzyeh Ghassemi PhD '17, PD '17, who is now an assistant professor of computer science and medicine at the University of Toronto.

Forced feature-learning

For years, the MIT researchers have worked with the Center for Laryngeal Surgery and Voice Rehabilitation to develop and analyze data from a sensor to track subject voice usage during all waking hours. The sensor is an accelerometer with a node that sticks to the neck and is connected to a smartphone. As the person talks, the smartphone gathers data from the displacements in the accelerometer.

In their work, the researchers collected a week’s worth of this data — called “time-series” data — from 104 subjects, half of whom were diagnosed with vocal cord nodules. For each patient, there was also a matching control, meaning a healthy subject of similar age, sex, occupation, and other factors.

Traditionally, experts would need to manually identify features that may be useful for a model to detect various diseases or conditions. That helps prevent a common machine-learning problem in health care: overfitting. That’s when, in training, a model “memorizes” subject data instead of learning just the clinically relevant features. In testing, those models often fail to discern similar patterns in previously unseen subjects.

“Instead of learning features that are clinically significant, a model sees patterns and says, ‘This is Sarah, and I know Sarah is healthy, and this is Peter, who has a vocal cord nodule.’ So, it’s just memorizing patterns of subjects. Then, when it sees data from Andrew, which has a new vocal usage pattern, it can’t figure out if those patterns match a classification,” Gonzalez Ortiz says.

The main challenge, then, was preventing overfitting while automating manual feature engineering. To that end, the researchers forced the model to learn features without subject information. For their task, that meant capturing all moments when subjects speak and the intensity of their voices.

As their model crawls through a subject’s data, it’s programmed to locate voicing segments, which comprise only roughly 10 percent of the data. For each of these voicing windows, the model computes a spectrogram, a visual representation of the spectrum of frequencies varying over time, which is often used for speech processing tasks. The spectrograms are then stored as large matrices of thousands of values.

But those matrices are huge and difficult to process. So, an autoencoder — a neural network optimized to generate efficient data encodings from large amounts of data — first compresses the spectrogram into an encoding of 30 values. It then decompresses that encoding into a separate spectrogram.  

Basically, the model must ensure that the decompressed spectrogram closely resembles the original spectrogram input. In doing so, it’s forced to learn the compressed representation of every spectrogram segment input over each subject’s entire time-series data. The compressed representations are the features that help train machine-learning models to make predictions.  

Mapping normal and abnormal features

In training, the model learns to map those features to “patients” or “controls.” Patients will have more voicing patterns than will controls. In testing on previously unseen subjects, the model similarly condenses all spectrogram segments into a reduced set of features. Then, it’s majority rules: If the subject has mostly abnormal voicing segments, they’re classified as patients; if they have mostly normal ones, they’re classified as controls.

In experiments, the model performed as accurately as state-of-the-art models that require manual feature engineering. Importantly, the researchers’ model performed accurately in both training and testing, indicating it’s learning clinically relevant patterns from the data, not subject-specific information.

Next, the researchers want to monitor how various treatments — such as surgery and vocal therapy — impact vocal behavior. If patients’ behaviors move form abnormal to normal over time, they’re most likely improving. They also hope to use a similar technique on electrocardiogram data, which is used to track muscular functions of the heart.