MIT today announced the launch of the MIT Intelligence Quest, an initiative to discover the foundations of human intelligence and drive the development of technological tools that can positively influence virtually every aspect of society.

The announcement was first made in a letter MIT President L. Rafael Reif sent to the Institute community.

At a time of rapid advances in intelligence research across many disciplines, the Intelligence Quest — MIT IQ— will encourage researchers to investigate the societal implications of their work as they pursue hard problems lying beyond the current horizon of what is known.

Some of these advances may be foundational in nature, involving new insight into human intelligence, and new methods to allow machines to learn effectively. Others may be practical tools for use in a wide array of research endeavors, such as disease diagnosis, drug discovery, materials and manufacturing design, automated systems, synthetic biology, and finance.

“Today, we set out to answer two big questions," Reif says. “How does human intelligence work, in engineering terms? And how can we use that deep grasp of human intelligence to build wiser and more useful machines, to the benefit of society?”

MIT IQ: The Core and The Bridge

MIT is poised to lead this work through two linked entities within MIT IQ. One, “The Core,” will advance the science and engineering of both human and machine intelligence. A key output of this work will be machine-learning algorithms. At the same time, MIT IQ seeks to advance our understanding of human intelligence by using insights from computer science.

The second entity, “The Bridge,” will be dedicated to the application of MIT discoveries in natural and artificial intelligence to all disciplines, and it will host state-of-the-art tools from industry and research labs worldwide.

The Bridge will provide a variety of assets to the MIT community, including intelligence technologies, platforms, and infrastructure; education for students, faculty, and staff about AI tools; rich and unique data sets; technical support; and specialized hardware.

Along with developing and advancing the technologies of intelligence, MIT IQ researchers will also investigate the societal and ethical implications of advanced analytical and predictive tools. There are already active projects and groups at the Institute investigating autonomous systems, media and information quality, labor markets and the work of the future, innovation and the digital economy, and the role of AI in the legal system.

In all its activities, MIT IQ is intended to take advantage of — and strengthen — the Institute’s culture of collaboration. MIT IQ will connect and amplify existing excellence across labs and centers already engaged in intelligence research. It will also establish shared, central spaces conducive to group work, and its resources will directly support research.

“Our quest is meant to power world-changing possibilities,” says Anantha Chandrakasan, dean of the MIT School of Engineering and Vannevar Bush Professor of Electrical Engineering and Computer Science. Chandrakasan, in collaboration with Provost Martin Schmidt and all four of MIT’s other school deans, has led the development and establishment of MIT IQ.

“We imagine preventing deaths from cancer by using deep learning for early detection and personalized treatment,” Chandrakasan continues. “We imagine artificial intelligence in sync with, complementing, and assisting our own intelligence. And we imagine every scientist and engineer having access to human-intelligence-inspired algorithms that open new avenues of discovery in their fields. Researchers across our campus want to push the boundaries of what’s possible.”

Engaging energetically with partners

In order to power MIT IQ and achieve results that are consistent with its ambitions, the Institute will raise financial support through corporate sponsorship and philanthropic giving.

MIT IQ will build on the model that was established with the MIT–IBM Watson AI Lab, which was announced in September 2017. MIT researchers will collaborate with each other and with industry on challenges that range in scale from the very broad to the very specific.

“In the short time since we began our collaboration with IBM, the lab has garnered tremendous interest inside and outside MIT, and it will be a vital part of MIT IQ,” Reif says.

John E. Kelly III, IBM senior vice president for cognitive solutions and research, says: “To take on the world’s greatest challenges and seize its biggest opportunities, we need to rapidly advance both AI technology and our understanding of human intelligence. Building on decades of collaboration — including our extensive joint MIT–IBM Watson AI Lab — IBM and MIT will together shape a new agenda for intelligence research and its applications. We are proud to be a cornerstone of this expanded initiative.”

MIT will seek to establish additional entities within MIT IQ, in partnership with corporate and philanthropic organizations.

Why MIT

MIT has been on the frontier of intelligence research since the 1950s, when pioneers Marvin Minsky and John McCarthy helped establish the field of artificial intelligence.

MIT now has over 200 principal investigators whose research bears directly on intelligence. Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT Department of Brain and Cognitive Sciences (BCS) — along with the McGovern Institute for Brain Research and the Picower Institute for Learning and Memory — collaborate on a range of projects. MIT is also home to the National Science Foundation–funded center for Brains, Minds and Machines (CBMM) — the only national center of its kind.

Faculty from across the Institute will participate in the initiative, including researchers in the Media Lab, the Operations Research Center, the Institute for Data, Systems, and Society, the Sloan School of Management, the School of Architecture and Planning, and the School of Humanities, Arts, and Social Sciences.

“Our quest will amount to a journey taken together by all five schools at MIT,” Schmidt says. “Success will rest on a shared sense of purpose and a mix of contributions from a wide variety of disciplines. I’m excited by the new thinking we can help unlock.”

At the heart of MIT IQ will be collaboration among researchers in human and artificial intelligence.

“To revolutionize the field of artificial intelligence, we should continue to look to the roots of intelligence: the brain,” says James DiCarlo, department head and Peter de Florez Professor of Neuroscience in the Department of Brain and Cognitive Sciences. “By working with engineers and artificial intelligence researchers, human intelligence researchers can build models of the brain systems that produce intelligent behavior. The time is now, as model building at the scale of those brain systems is now possible. Discovering how the brain works in the language of engineers will not only lead to transformative AI — it will also illuminate entirely new ways to repair, educate, and augment our own minds.”

Daniela Rus, the Andrew (1956) and Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT, and director of CSAIL, agrees. MIT researchers, she says, “have contributed pioneering and visionary solutions for intelligence since the beginning of the field, and are excited to make big leaps to understand human intelligence and to engineer significantly more capable intelligent machines. Understanding intelligence will give us the knowledge to understand ourselves and to create machines that will support us with cognitive and physical work.”

David Siegel, who earned a PhD in computer science at MIT in 1991 pursuing research at MIT’s Artificial Intelligence Laboratory, and who is a member of the MIT Corporation and an advisor to the MIT Center for Brains, Minds, and Machines, has been integral to the vision and formation of MIT IQ and will continue to help shape the effort. “Understanding human intelligence is one of the greatest scientific challenges,” he says, “one that helps us understand who we are while meaningfully advancing the field of artificial intelligence.” Siegel is co-chairman and a founder of Two Sigma Investments, LP.

The fruits of research

MIT IQ will thus provide a platform for long-term research, encouraging the foundational advances of the future. At the same time, MIT professors and researchers may develop technologies with near-term value, leading to new kinds of collaborations with existing companies — and to new companies.

Some such entrepreneurial efforts could be supported by The Engine, an Institute initiative launched in October 2016 to support startup companies pursuing particularly ambitious goals.

Other innovations stemming from MIT IQ could be absorbed into the innovation ecosystem surrounding the Institute — in Kendall Square, Cambridge, and the Boston metropolitan area. MIT is located in close proximity to a world-leading nexus of biotechnology and medical-device research and development, as well as a cluster of leading-edge technology firms that study and deploy machine intelligence. 

MIT also has roots in centers of innovation elsewhere in the United States and around the world, through faculty research projects, institutional and industry collaborations, and the activities and leadership of its alumni. MIT IQ will seek to connect to innovative companies and individuals who share MIT’s passion for work in intelligence.

Eric Schmidt, former executive chairman of Alphabet, has helped MIT form the vision for MIT IQ. “Imagine the good that can be done by putting novel machine-learning tools in the hands of those who can make great use of them,” he says. “MIT IQ can become a fount of exciting new capabilities.”

“I am thrilled by today’s news,” Reif says. “Drawing on MIT’s deep strengths and signature values, culture, and history, MIT IQ promises to make important contributions to understanding the nature of intelligence, and to harnessing it to make a better world.”

“MIT is placing a bet,” he says, “on the central importance of intelligence research to meeting the needs of humanity.”

The field of 3-D printing has come a long way since the first rapid prototyping patent was rejected in 1980. The technology has evolved from basic designs to a wide range of highly-customizable objects. Still, there’s a big issue: Once objects are printed, they’re final. If you need a change, you’ll need a reprint.

But imagine if that weren’t the case — if, for example, you could change the color of your smartphone case or earrings on demand.

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have gotten closer to making that a reality. In a new paper, they present ColorMod, a method for repeatedly changing the colors of 3-D printed objects, after fabrication.

Using their own 3-D printable ink that changes color when exposed to ultraviolet light, the team can recolor a multicolored object in just over 20 minutes — and they say they expect that number to decrease significantly with future improvements.

While the project is currently focused on plastics and other common 3-D printing materials, the researchers say that eventually people could instantly change the color of their clothes and other items.

“Largely speaking, people are consuming a lot more now than 20 years ago, and they’re creating a lot of waste,” says Stefanie Mueller, the X-Consortium Career Development Assistant Professor in the departments of Electrical Engineering and Computer Science and Mechanical Engineering. “By changing an object’s color, you don’t have to create a whole new object every time.”

Mueller co-authored the paper with postdoc Parinya Punpongsanon, undergraduate Xin Wen, and researcher David Kim. It has been accepted to the 2018 Conference on Human Factors in Computing Systems, an annual Association for Computing Machinery (ACM) Special Interest Group on Computer-Human Interaction (SIGCHI) event being held in April in Montreal.

How it works

Previous color-changing systems have been somewhat limited in their capabilities, using single colors and 2-D designs, for example.

To move beyond single-color systems, the team developed a simple hardware/software workflow. First, using the ColorMod interface, users upload their 3-D model, pick their desired color patterns, and then print their fully colored object.  

After printing, changing the multicolored objects involves using ultraviolet light to activate desired colors and visible light to deactivate others. Specifically, the team uses an ultraviolet light to change the pixels on an object from transparent to colored, and a regular office projector to turn them from colored to transparent.

The team’s custom ink is made of a base dye, a photoinitiator, and light-adaptable dyes. The light-adaptable (photochromic) dyes bring out the color in the base dye, and the photoinitiator lets the base dye harden during 3-D printing.

“Appearance adaptivity in general is always a superior feature to have, and we’ve seen many other kinds of adaptivity enabled with manufactured objects,” says Changxi Zheng, an associate professor at Columbia University who co-directs Columbia’s Computer Graphics Group. “This work is a true breakthrough in being able to change the color of objects without repainting them.”

The team tested ColorMod on three criteria: recoloring time, precision, and how quickly the color decayed. A full recoloring process took 23 minutes. However, the researchers note that they could speed up the process by using a more powerful light or adding more light-adaptable dye to the ink.

They also found the colors to be a bit grainy, which they hope to improve on by activating colors closer together on an object. For example, activating blue and red might show purple, while activating red and green would show yellow.

Mueller says that the goal is for people to be able to rapidly match their accessories to their outfits in an efficient, less wasteful way. Another idea is for retail stores to be able to customize products in real-time, if, for example, a shopper wants to try on an article of clothing or accessory in a different color.

“This is the first 3-D-printable photochromic system that has a complete printing and recoloring process that’s relatively easy for users,” Punpongsanon says. “It’s a big step for 3-D printing to be able to dynamically update the printed object after fabrication in a cost-effective manner.”

Halfway through his sophomore year, Liang Zhou made a decision that changed the course of his academic career. The electrical engineering and computer science major enrolled in three classes in brain and cognitive sciences, an area he hadn’t studied before. Though he can’t pinpoint his exact reasoning for doing so at the time — perhaps it was intuition —  he has no regrets.

As he begins his last semester at MIT, Zhou is “really, really glad” he made that choice, and is now completing a double major in electrical engineering and computer science and brain and cognitive sciences.

“I’m interested in studying computer science because of its wealth of applications in other domains,” Zhou says. Take neuroscience: “It’s about the brain and it’s about how we as people fundamentally think about things, so understanding others. It’s a way to connect both the very hard science and rigor to something that’s very relevant to everybody.”

As he prepares to pursue a master’s degree in computational neuroscience at the University College London as a Marshall Scholar, the native of Riverside, California, can clearly articulate his research motivation: “I am fascinated by how people work, why we do what we do, and why we think what we think.”

Intuitive physics

For Zhou, his intense introduction to neuroscience marked “a paradigm shift.” While he began to study the brain in his classes, Zhou also opted to pursue research in a field where he could apply his coding skills. He found a great fit: the Computational Cognitive Science Laboratory, led by Professor Josh Tenenbaum. The lab group uses a combination of computer modeling and behavioral experiments to understand the basis of human learning.

Under the mentorship of Tobias Gerstenberg, a postdoc in the Tenenbaum lab, Zhou  has worked on projects that investigate how people perceive their environments.

Zhou began with a research project that asked subjects to assess the structural stability of a brick tower. Subjects were asked what would happen to the brick tower if certain blocks were removed, and to assess how important single bricks were for the stability of the overall structure. “We calculated responsibilities for different brick configurations and compared them to people’s assessments, which gave us a better notion for people’s judgements about stability,” Zhou says.

What fascinated Zhou most was the disconnect between the subjects’ judgements and the ground truth. “What people think will happen is usually not what happens in terms of those bricks, so we created a model … which was better in line with people’s predictions and not the ground truth.” That model, called the hypothetical simulation model, was detailed in a conference paper for the 39th annual meeting of the Cognitive Science Society, which was held this past July.

That model and the disconnect it details led Zhou to his current research interest: understanding intuitive physics.

“When you take an apple and you drop it, you don’t think F=ma, therefore it’s going to drop with one Newton and it’s going to hit the ground at this time,” Zhou says. “You think, ‘Oh, it’s an apple and it’s going to fall.’ We have an intuition for how the world works, and we have a mental notion of how one physical event can cause others to happen … but we never explicitly learn this material, and we learn it so readily and so easily.”

Zhou wants to investigate how that intuition arises from neurons, and that begins with the dedicated study he plans to do in graduate school: “Computational neuroscience is a lot about modeling the actual neurons themselves. … In the future, I want to do research in cognitive neuroscience. I think it’s very important to have a very solid statistical and mathematical background, and I hope that studying computational neuroscience will give me that.” 

Understanding impacts

In the future, Zhou sees himself continuing his research and sharing its applications.

“There’s no better way to actually contribute to this field of research than to be in academia and to do research, but I also realize that research isn’t particularly useful if it’s not applied to something,” Zhou says. “With this line of research, there are so many applications in society. With a better understanding of how we think … we could have a better sense for what it means to understand somebody.”

Zhou also cites recent research in behavioral economics that highlights the surprising strength of irrational thinking. “We assume people are rational, and we assume people are logical, but they are really not. And [knowing] that helps you create better models of how people work.”

He envisions these types of models informing future education and public policy. “This research isn’t going anywhere if we don’t actually put it somewhere,” Zhou says. To share his vision, Zhou also hopes to become an active voice in science policy and public science advocacy. However, he points out that’s far in the future: “For the moment I just want to do a PhD for more in-depth and hands-on research experience, but I’d love to get involved in making this have tangible effects on the world in the future if there’s a path there for me.”

During his time at MIT, Zhou has also interned in software engineering at NextJump, Lucid Software, and Apple. Zhou served at conference chair for the 2017 EECScon, a U.S. undergraduate-led research conference. In addition to his research in the Tenenbaum group, he also performs computational neuroscience research at Harvard University, and has undertaken research projects in the MIT Media Lab and the Computer Science and Artificial Intelligence Laboratory through the Undergraduate Research Opportunities Program (UROP).

Zhou is an active member of the MIT Asian Dance Team, where he has served on the executive board for two years. He has also choreographed hip-hop pieces for MIT DanceTroupe. Additionally, Zhou served as teaching assistant for both undergraduate- and graduate-level machine learning courses for MIT students. This past year, Zhou was awarded the Hans Lukas Teuber Award for Outstanding Academics.

For related articles, please visit the MIT News website.

Image: Courtesy of TetraScience

Rob Matheson | MIT News Office

Although Internet-connected “smart” devices have in recent years penetrated numerous industries and private homes, the technological phenomenon has left the research lab largely untouched. Spreadsheets, individual software programs, and even pens and paper remain standard tools for recording and sharing data in academic and industry labs.

TetraScience, co-founded by Spin Wang SM ’15, a graduate of electrical engineering and computer science, has developed a data-integration platform that connects disparate types of lab equipment and software systems, in-house and at outsourced drug developers and manufacturers. It then unites the data from all these sources in the cloud for speedier and more accurate research, cost savings, and other benefits.

“Software and hardware systems [in labs] cannot communicate with each other in a consistent way,” says Wang, TetraScience’s chief technology officer, who co-founded the startup with former Harvard University postdocs Salvatore Savo and Alok Tayi. “Data flows through systems in a very fragmented manner and there are a lot of siloed data sets [created] in the life sciences. Humans must manually copy and paste information or write it down on paper, [which] is a lengthy manual process that’s error prone.”

TetraScience has developed an Internet of Things (IoT) hub that plugs into most lab equipment, including freezers, ovens, incubators, scales, pH meters, syringe pumps, and autoclaves. The hub can also continuously collect relevant data — such as humidity, temperature, gas concentration and oxygen levels, vibration, light intensity, and mass air flow — and shoot it to TetraScience’s centralized data-integration platform in the cloud. TetraScience also has custom integration methods for more complicated instruments and software.

In the cloud dashboard, researchers can monitor equipment in real time and set alerts if any equipment deviates from ideal conditions. Data appears as charts, graphs, percentages, and numbers — somewhat resembling the easily readable Google Analytics dashboard. Equipment can be tracked for usage and efficiency over time to determine if, say, a freezer is slowly warming and compromising samples. Researchers can also comb through scores of archived data, all located in one place.

“Our technology is establishing a ‘data highway’ system between different entities, software and hardware, within life sciences labs. We make facilitating data seamless, faster, more accurate, and more efficient,” says Wang, who was named to this year’s Forbes 30 Under 30 list of innovators for his work with TetraScience.

More than 70 major pharmaceutical and biotech firms, including many in Cambridge, Massachusetts, use the platform. Numerous labs at MIT and Harvard are users, as well.

Pain in the lab

For Wang and his TetraScience co-founders, building their smart solution was personal.

As a Cornell University undergraduate, Wang worked in the Cornell Semiconducting RF Lab on high-energy physics research. Frustrated by the time and effort required to manually record data, he developed his own system that connected and controlled more than 10 instruments, such as a signal generator, power meter, frequency counter, and power amplifier.

Years later, as an MIT master’s student studying microelectromechanical systems, Wang worked on sensing technologies and processing of radio frequency signals under the guidance of Professor Dana Weinstein, now at Purdue University. During his final year, he wound up at the MIT Media Lab, working on a 3-D printing project with Tayi, who had spent his academic career toiling away in materials science, chemistry, and other labs. Tayi and Savo were already conducting market research around potential opportunities for IoT in labs.

All three bonded over a shared dislike for data-collecting tools that have remained relatively unchanged in labs for a half-century. “We felt the pain of manually tracking data and not having a consistent interface for all our equipment,” Wang says.

This is especially troublesome at scale. Large pharmaceutical or biotechnology firms, for instance, can have several hundreds or thousands of instruments, all with different hardware running on different software. Humans must record data and input it manually into dozens of separate recording systems, which leads to errors. People also must be physically in a lab to control experiments. Smart labs were the new frontier, Wang, Savo, and Tayi agreed.

In 2014, the three launched TetraScience to build a platform that connected equipment and pooled data into a single place in the cloud — similar to the one Wang created at Cornell, but more advanced. Back then, they used a slightly modified Raspberry Pi as their “hub,” while they refined their software and hardware.

For early-stage startup advice, the startup turned to the Industrial Liaison Program and MIT’s Venture Mentoring Service, and leveraged MIT’s vast alumni network for feedback on their technology and business plan. “We definitely benefited from MIT,” Wang says.

Saving time and money

An early trial for the platform was with the Media Lab, where researchers used the platform to monitor not equipment, but beehives. The researchers were studying how hives could be implemented into building infrastructure and how design and materials could promote bee health. As bees are sensitive to changes in environment, the researchers needed to constantly monitor temperature and humidity around hives over several months, which would be challenging if done manually.

Using TetraScience’s platform, the researchers were captured all the necessary data for their project, without suiting up and approaching all the hives daily — saving “hundreds of hours … and 686 bee stings,” according to the startup. Testing at MIT, Wang says, “helped us gain an understanding of the industry and value proposition.”

From there, the TetraScience platform found its way into more biotech companies and into more than 60 percent of the world’s top 20 pharmaceutical companies, according to the startup. Benefits of today’s TetraScience platform include speeding up research, improving compliance, producing better-quality data and, ultimately, saving millions of dollars and countless hours of work, Wang says.

Numerous case studies, listed on the startup’s webpage, showcase the platform’s efficacy and value at major pharmaceutical firms and cancer research centers, and at Harvard and MIT.

For example, in the final stages of approval of a multibillion-dollar drug, a large pharmaceutical firm conducted an accelerated lifetime test, where any prolonged deviation from preset conditions would require restarting the experiment, at the cost of millions of dollars, weeks of unusable data, and delayed commercialization. Within a few weeks of the test’s conclusion, a major deviation in one experiment occurred late at night. Within seconds, according to the study, TetraScience’s platform detected the deviation and alerted scientists, who caught it immediately, stopping any significant damage.

The platform also offers benefits for determining equipment efficiency and usage. In a 2017 case study with another pharmaceutical firm, TetraScience monitored 70 pieces of equipment. The startup flagged 23 instruments as “heavily underused.” The firm used that data to reduce service contracts for 14 instruments and sell nine instruments, leading to improved efficiency and hundreds of thousands of dollars in savings that could be put toward more research and development. 

Although the startup’s focus is on pharmaceutical and biotechnology industries, the platform could also be used in oil and gas, brewing, and food and chemistry industries to see similar benefits. “Those industries all use similar instruments [as life science labs] and produce the same kind of data, such as monitoring the pH of beer, so we will get into those industries in the future,” Wang says.

Next, the researchers use machine-learning techniques to determine correlations between specific values for the variables in the formulae and the measured properties of the resulting microstructures. This gives the system a rigorous way to translate back and forth between microstructures and their properties.

On automatic

Every step in this process, Matusik emphasizes, is completely automated, including the measurement of similarities, the clustering, the skeleton extraction, the formula derivation, and the correlation of geometries and properties. As such, the approach would apply as well to any collection of microstructures evaluated according to any criteria.

By the same token, Matusik explains, the MIT researchers’ system could be used in conjunction with existing approaches to materials design. Besides taking inspiration from biological materials, he says, researchers will also attempt to design microstructures by hand. But either approach could be used as the starting point for the sort of principled exploration of design possibilities that the researchers’ system affords.

“You can throw this into the bucket for your sampler,” Matusik says. “So we guarantee that we are at least as good as anything else that has been done before.”

In the new paper, the researchers do report one aspect of their analysis that was not automated: the identification of the physical mechanisms that determine the microstructures’ properties. Once they had the skeletons of several different families of microstructures, they could determine how those skeletons would respond to physical forces applied at different angles and locations.

But even this analysis is subject to automation, Chen says. The simulation software that determines the microstructures’ properties can also identify the structural elements that deform most under physical pressure, a good indication that they play an important functional role.

The work was supported by the U.S. Defense Advanced Research Projects Agency’s Simplifying Complexity in Scientific Discovery program.

Most recent advances in artificial-intelligence systems such as speech- or face-recognition programs have come courtesy of neural networks, densely interconnected meshes of simple information processors that learn to perform tasks by analyzing huge sets of training data.

But neural nets are large, and their computations are energy intensive, so they’re not very practical for handheld devices. Most smartphone apps that rely on neural nets simply upload data to internet servers, which process it and send the results back to the phone.

Now, MIT researchers, led by EECS, have developed a special-purpose chip that increases the speed of neural-network computations by three to seven times over its predecessors, while reducing power consumption 94 to 95 percent. That could make it practical to run neural networks locally on smartphones or even to embed them in household appliances.

“The general processor model is that there is a memory in some part of the chip, and there is a processor in another part of the chip, and you move the data back and forth between them when you do these computations,” says Avishek Biswas, an MIT graduate student in electrical engineering and computer science, who led the new chip’s development.

“Since these machine-learning algorithms need so many computations, this transferring back and forth of data is the dominant portion of the energy consumption. But the computation these algorithms do can be simplified to one specific operation, called the dot product. Our approach was, can we implement this dot-product functionality inside the memory so that you don’t need to transfer this data back and forth?”

Biswas and his thesis advisor, Anantha Chandrakasan, dean of MIT’s School of Engineering and the Vannevar Bush Professor of Electrical Engineering and Computer Science, describe the new chip in a paper that Biswas is presenting this week at the International Solid State Circuits Conference.

Back to analog

Neural networks are typically arranged into layers. A single processing node in one layer of the network will generally receive data from several nodes in the layer below and pass data to several nodes in the layer above. Each connection between nodes has its own “weight,” which indicates how large a role the output of one node will play in the computation performed by the next. Training the network is a matter of setting those weights.

A node receiving data from multiple nodes in the layer below will multiply each input by the weight of the corresponding connection and sum the results. That operation — the summation of multiplications — is the definition of a dot product. If the dot product exceeds some threshold value, the node will transmit it to nodes in the next layer, over connections with their own weights.

A neural net is an abstraction: The “nodes” are just weights stored in a computer’s memory. Calculating a dot product usually involves fetching a weight from memory, fetching the associated data item, multiplying the two, storing the result somewhere, and then repeating the operation for every input to a node. Given that a neural net will have thousands or even millions of nodes, that’s a lot of data to move around.

But that sequence of operations is just a digital approximation of what happens in the brain, where signals traveling along multiple neurons meet at a “synapse,” or a gap between bundles of neurons. The neurons’ firing rates and the electrochemical signals that cross the synapse correspond to the data values and weights. The MIT researchers’ new chip improves efficiency by replicating the brain more faithfully.

In the chip, a node’s input values are converted into electrical voltages and then multiplied by the appropriate weights. Only the combined voltages are converted back into a digital representation and stored for further processing.

The chip can thus calculate dot products for multiple nodes — 16 at a time, in the prototype — in a single step, instead of shuttling between a processor and memory for every computation.

All or nothing

One of the keys to the system is that all the weights are either 1 or -1. That means that they can be implemented within the memory itself as simple switches that either close a circuit or leave it open. Recent theoretical work suggests that neural nets trained with only two weights should lose little accuracy — somewhere between 1 and 2 percent.

Biswas and Chandrakasan’s research bears that prediction out. In experiments, they ran the full implementation of a neural network on a conventional computer and the binary-weight equivalent on their chip. Their chip’s results were generally within 2 to 3 percent of the conventional network’s.

"This is a promising real-world demonstration of SRAM-based in-memory analog computing for deep-learning applications,” says Dario Gil, vice president of artificial intelligence at IBM. "The results show impressive specifications for the energy-efficient implementation of convolution operations with memory arrays. It certainly will open the possibility to employ more complex convolutional neural networks for image and video classifications in IoT [the internet of things] in the future."

Today, most web browsers have private-browsing modes, in which they temporarily desist from recording the user’s browsing history.

But data accessed during private browsing sessions can still end up tucked away in a computer’s memory, where a sufficiently motivated attacker could retrieve it.

This week, at the Network and Distributed Systems Security Symposium, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Harvard University presented a paper describing a new system, dubbed Veil, that makes private browsing more private.

Veil would provide added protections to people using shared computers in offices, hotel business centers, or university computing centers, and it can be used in conjunction with existing private-browsing systems and with anonymity networks such as Tor, which was designed to protect the identity of web users living under repressive regimes.

“Veil was motivated by all this research that was done previously in the security community that said, ‘Private-browsing modes are leaky — Here are 10 different ways that they leak,’” says Frank Wang, an MIT graduate student in electrical engineering and computer science and first author on the paper. “We asked, ‘What is the fundamental problem?’ And the fundamental problem is that [the browser] collects this information, and then the browser does its best effort to fix it. But at the end of the day, no matter what the browser’s best effort is, it still collects it. We might as well not collect that information in the first place.”

Wang is joined on the paper by his two thesis advisors: Nickolai Zeldovich, an associate professor of electrical engineering and computer science at MIT, and James Mickens, an associate professor of computer science at Harvard.

Shell game

With existing private-browsing sessions, Wang explains, a browser will retrieve data much as it always does and load it into memory. When the session is over, it attempts to erase whatever it retrieved.

But in today’s computers, memory management is a complex process, with data continuously moving around between different cores (processing units) and caches (local, high-speed memory banks). When memory banks fill up, the operating system might transfer data to the computer’s hard drive, where it could remain for days, even after it’s no longer being used.

Generally, a browser won’t know where the data it downloaded has ended up. Even if it did, it wouldn’t necessarily have authorization from the operating system to delete it.

Veil gets around this problem by ensuring that any data the browser loads into memory remains encrypted until it’s actually displayed on-screen. Rather than typing a URL into the browser’s address bar, the Veil user goes to the Veil website and enters the URL there. A special server — which the researchers call a blinding server — transmits a version of the requested page that’s been translated into the Veil format.

The Veil page looks like an ordinary webpage: Any browser can load it. But embedded in the page is a bit of code — much like the embedded code that would, say, run a video or display a list of recent headlines in an ordinary page — that executes a decryption algorithm. The data associated with the page is unintelligible until it passes through that algorithm.

Decoys

Once the data is decrypted, it will need to be loaded in memory for as long as it’s displayed on-screen. That type of temporarily stored data is less likely to be traceable after the browser session is over. But to further confound would-be attackers, Veil includes a few other security features.

One is that the blinding servers randomly add a bunch of meaningless code to every page they serve. That code doesn’t affect the way a page looks to the user, but it drastically changes the appearance of the underlying source file. No two transmissions of a page served by a blinding sever look alike, and an adversary who managed to recover a few stray snippets of decrypted code after a Veil session probably wouldn’t be able to determine what page the user had visited.

If the combination of run-time decryption and code obfuscation doesn’t give the user an adequate sense of security, Veil offers an even harder-to-hack option. With this option, the blinding server opens the requested page itself and takes a picture of it. Only the picture is sent to the Veil user, so no executable code ever ends up in the user’s computer. If the user clicks on some part of the image, the browser records the location of the click and sends it to the blinding server, which processes it and returns an image of the updated page.

The back end

Veil does, of course, require web developers to create Veil versions of their sites. But Wang and his colleagues have designed a compiler that performs this conversion automatically. The prototype of the compiler even uploads the converted site to a blinding server. The developer simply feeds the existing content for his or her site to the compiler.

A slightly more demanding requirement is the maintenance of the blinding servers. These could be hosted by either a network of private volunteers or a for-profit company. But site managers may wish to host Veil-enabled versions of their sites themselves. For web services that already emphasize the privacy protections they afford their customers, the added protections provided by Veil could offer a competitive advantage.

“Veil attempts to provide a private browsing mode without relying on browsers,” says Taesoo Kim, an assistant professor of computer science at Georgia Tech, who was not involved in the research. “Even if end users didn't explicitly enable the private browsing mode, they still can get benefits from Veil-enabled websites. Veil aims to be practical — it doesn't require any modification on the browser side — and to be stronger — taking care of other corner cases that browsers do not have full control of.”