T e c h

T oday

THInk Of THe POssIbIlITIes

Medical micro-implants could work wonders. cIMM, a new research center, tackles the engineering work.
By MIke Vargo

10

engineering news

One of humankinds oldest fantasies is being able to control the external world by thought. That fantasy is now moving toward reality. In St. Louis, where a teenaged epilepsy patient had an electronic interface placed in his skull to monitor seizures, researchers found that the boy could use the interface to play the video game Space Invaders just by thinking about the moves he wanted to make. In a laboratory at the University of Pittsburgh, a similarly wired monkey moved a robotic arm to reach for food and feed itself, just by thinking about it. The research field is called BCI (for brain-computer interfaces) and one can easily think of practical uses. Most obviously, people with spinal cord injuries or disabling diseases could someday be fitted with systems that give them vastly more capability and freedom. For many such people, says Carnegie Mellon Professor Jeyanandh Paramesh, the area of the brain that controls the movement of limbs is still working the brain is able to say okay, move this hand but the link is broken. The general aim of the research is to replace that broken pathway. Paramesh, of Electrical and Computer Engineering, is one of a dozen CIT faculty affiliated with Carnegie Mellons new Center for Implantable Medical Microsystems (CIMM). The center is already perhaps the largest of its kind at a university without a medical school. CIMM faculty collaborate with medical researchers across Pittsburgh at Pitt, at West Penn and Allegheny General hospitals and with other researchers at Carnegie Mellon (for instance, the universitys Center for the Neural Basis of Cognition is world-renowned in cognitive science). A key focus area at CIMM is developing the hardware and software needed to build next-generation brain-computer interfaces and make them more useful. This research is urgently needed, according to ECE Professor Gary Fedder, who is director of both CIMM and its parent entity ICES, the Institute for Complex Engineered Systems. A lot of good work is being done in understanding how the brain and body operate, Fedder says. Now there have to be more people doing the engineered systems part of the work, to complement that. Fedder can draw from top-flight talent and hes been recruiting as well. The stories of two faculty members, one new and one already on board, illustrate the scope of the engineering challenges.

Getting Into (and Out of) Ones Head

Brains are much more complex than computers but they are electrical, with specific neurons firing for specific tasks and neurons for motor activity are in the outer layer, the cortex. Direct interaction between brains and computer-based machines is thus possible by capturing and processing the relevant signals. But to do this one must make sense of the morass of activity generated by a brain, figuring out which signals mean what. That is where Byron Yu comes in. Yu has joined the ECE faculty after postdoctoral research at Stanford. He explains that most current studies of brain activity, in both human and animal subjects, are done by recording from one neuron at a time, using a single electrode in the cortex. You

cooL, reaL-TIMe aNd BIodegradaBLe

can learn some things that way, he goes on. Say you have the subject reach to the right twenty times, then to the left twenty times. Just by looking at one neuron at a time you can get the equivalent of yes-no answers: right or left? In recent years, says Yu, experimenters have begun to use multi-electrode arrays to monitor tens to hundreds of neurons simultaneously. Now they can look for patterns of neural activity that drive more complex tasks like grasping and manipulating objects. But the signal-to-noise quality is not as good with the arrays, he notes. And now we are collecting humongous data sets every day that can be leveraged for furthering our understanding of the brain and ultimately helping human patients. We could be doing much more with the data than we are; we dont have the mathematical tools that we had for studying one neuron at a time. Yus specialty is developing and applying mathematical algorithms for studying large-scale neural activity. As for why he has come to Carnegie Mellon, he says, This work is interdisciplinary. It involves neuroscience, signal processing, statistics, machine learning and other fields. I dont know of any place that has as many experts in each. Brain-computer work will require new hardware, too. Typical electrode arrays have tiny needles that penetrate into the cortex. Though not directly harmful, they are invasive, and eventually the brain forms scar tissue around them so the signals die out, says Jeyanandh Paramesh. Thats okay for short-term experiments but not for longterm therapeutic use. When you put something into a persons skull you want to leave it there and know its functioning. Paramesh is working with scientists at Pitt who have a planar array that would go under the skull, but sit on the surface of the brain. He is developing the hardware to go with it. The challenges are many, he says, because for long-term use the whole implanted unit has to run wirelessly: You dont want wires sticking through your head. Power is one problem, since onboard batteries would run down. So the implant will use a method called resonant inductive coupling. On the outside of the patients head, a tiny power unit would carry the primary coil of a micro-transformer. The secondary coil (which picks up induced voltage from the primary) would reside on the implant. Then comes transmitting the neural signals gathered by the electrodes. Shortrange wireless transmission may seem no big deal. But here again the amount of signal data is huge, and as Paramesh notes, you cant simply pump it all out to be processed externally. It would take too much bandwidth. It would consume too much power and create heat; the temperature rise would damage brain tissue. Circuitry is needed to compress and condition data before sending, Paramesh says. He has now designed a system that can sample and process data from any four of the arrays 32 electrodes at once. Hes trying to push that limit while balancing a host of further design problems, from keeping the overall form-factor tight to dealing with fabrication issues. Work of this type is incredibly complex. It will tax many skills, such as Carnegie Mellons well-known strengths in MEMS (micro-electro-mechanical systems, the manufacture of infinitesimal on-chip components) and yet, the work is moving forward.

among the coolest items in development at cIMM is a temperature sensor the size of a grain of uncooked rice. Its for use with cryosurgery, the injection of a super-cold fluid such as liquid nitrogen to kill tumors or unwanted tissue. Prostate cancer is often treated by cryosurgery. It has been successful, but there can be very bad outcomes if you freeze more tissue than you wanted, or freeze the wrong tissue. reducing that risk is the goal, says cIMM director gary Fedder, a co-PI on the project. currently, surgeons insert cryoprobe needles and apply the coolant with the aid of ultrasound imaging. What could help, Fedder says, is a system for more precise feedback on the spread of the freeze front. Meche Professor yoed rabin, an expert in cryoengineering, came up with the idea of the mini-sensors. Twenty or so would be implanted in and around the prostate before surgery. They would then sense and wirelessly transmit the temperatures of nearby tissue, to help assure that tumors are eliminated while vital functions are not. after surgery the sensors would biodegrade, leaving just harmless traces of microelectronic components. In fact, research on biodegradable implants is a field of its own at cIMM. The work here is led by Lee Weiss and Phil campbell, both of Biomedical engineering and computer Science; one application theyre pursuing is an implant to induce bone regeneration after severe injuries. Meanwhile, research scientist alan rosenbloom is looking into other types of micro-sensing. If you are a patient in a critical care unit, some of your vital signs, such as heart rate, can and will be monitored constantly. But others can only be checked by drawing periodic samples notably glucose level in the blood, which must be in the normal range to avoid glycemic shock. So rosenbloom is working on a real-time sensing device (technically, a microdialysis device) that can be inserted into the bloodstream on a needlepoint. The ultimate vision is having such sensors implanted long term in outpatients, to monitor not only blood sugar but the immune agents called cytokines. Which might pay big dividends: cytokine levels can be early markers of cancer.

11

What lies Ahead

Gary Fedder says that all biomedical engineering at CIMM has three goals: Make things small. Make them near-zero-invasive. And focus on modular engineering so we have pieces we can re-use and dont have to build a custom solution for every medical need. Looking ahead he sees myriad benefits flowing from CIMMs research. Some of the brain-computer work, for instance, could also be applied for inputting signals to the brain: to suppress tremors and seizures in people with neural disorders to improve the now-primitive retinal implant systems for giving sight to the blind. Or someday, maybe I could just think about this story and a computer would write it, Fedder smiles. But these things are far down the road. We are in seed stage, and were building.

Tech Today

T e c h

T o d ay

DOnT blAMe THe UseRs

Instead lets make computer security There is a growing consensus that the weakmore usable, says the head of a new est link in computer and Internet security is not Ph.d. program. vulnerable software, but user behavior. Every year,
according to various surveys, surprising numbers of users still fall for phishing scams. Others download things they shouldnt or disregard the most basic good practices, often wreaking havoc across their employers networks. Where Lorrie Cranor differs from many experts is in refusing to believe that user lapses are inevitable by-products of the ignorance of the masses. Most security breaches can be attributed to human error, she says, then quickly adds the punch line: which means they come from the failure of systems designers to meet human needs and accommodate human capacities and limitations. Cranor is one of the founders of an emerging research field called usable privacy and security. At Carnegie Mellon where she holds cross-appointments in Computer science and Engineering and Public Policy she directs one of the few comprehensive research centers in the field, CUPS: the CyLab Usable Privacy and Security Laboratory. Now she and her colleagues are building the worlds first usable privacy and security Ph.D. program. The CUPS Doctoral Training Program is being launched with a five-year, $3 million grant from the National Science Foundation. A charter class of six students enrolled in the fall of 09 and the program will take about 10 more each year. To grasp what students learn and do in this little-known field, lets join Lorrie Cranor for a whirlwind intro.
By MIke Vargo

Usable Privacy and security 101

Privacy and security are related but distinct issues. Clearly a website or computer must be secure (safe from malicious hackers) in order to give you privacy (control over what others can learn about you). On the other hand, Cranor points out, a site that you visit may have good security yet offer little privacy. The sites owners might be selling customer data to telemarketers, for instance. Research has shown that most users want privacy and security but arent sure how to get either. The goal of usability work, in a nutshell, is to maximize their chances. Cranor says there are three main strategies for doing so, one being to take obvious

1212

A Ph.D. STUDENT STAKES OUT THE HOME FRONT In the newly launched Ph.D. program for usable privacy and security, Michelle Mazurek is the Home Storage Student. She is working with partners including her faculty advisors, ECE professors Lujo Bauer and Greg Ganger, on new systems for managing digital storage in households. That is a growing problem. As Mazurek notes, many of us have text and multimedia files scattered across a multitude of devices at home: desktop PCs and laptops, video and music players and more. The project assumes that all devices can be wirelessly linked, and one nifty feature, Mazurek says, is a file-tagging system for seeing that all files go where theyre supposed to be, regardless of where theyre created. For instance, you can specify that all my work files go on the laptop in addition to the desktop. In a distributed environment of this type, privacy and security concerns loom large. Mazureks role is to address these, helping to develop methods to let each user in the home specify who can have access to what, under which conditions. As a first step, she and team members interviewed sample households to learn about needs and desires. Among the findings, Mazurek says, are that presence matters: people feel more comfortable with others seeing their files if they can be present to monitor it. And people want the ability to make ad-hoc access decisions instead of just setting policies a priori. Mazurek also learned that some people have strange habits. For instance, they try to hide sensitive files by giving them funny filenames; burying them in sub-folders. Its like burying valuables in the bottom of a drawer. The downside, of course, is that you can forget where you hid the gold watch and a persistent thief can still find it. But to Mazurek such things are more than amusing tidbits. What were seeing are unmet needs, or imperfectly met needs, for privacy and security, she says. Our job is to find better ways.

engineering news engineering news

decisions out of peoples hands by automating them. As she noted in a research paper, early antivirus programs prompted users to make a decision about every detected virus, whereas today the common default mode is to just delete or quarantine infected files. But many choices arent so conducive to automation. That leaves the other two approaches: designing features so theyre intuitive and easy to use and educating the users. Research by Cranor et al has shown that a great deal of confusion reigns. Many people conflate privacy and security. Some claim to be militant about privacy rights (theyre known as privacy fundamentalists), yet in experiments they will enter more personal data than needed for an online transaction. We all judge by appearances, and many judge a website to be trustworthy if it looks professionally done. The list goes on; solutions are needed.

fish stories and nutrition labels

Solution-wise, the CUPS research group has made perhaps its biggest splash thus far in user education. Have you seen Anti-Phishing Phil? Hes a cartoon character, a young fish, who was created by recent EPP Ph.D. graduate Steve Sheng and former Communication Design student Bryant Magnien. Phil and an older and wiser fish named PhishGuru star in an interactive, online game that teaches players how to recognize phishy emails and avoid getting hooked. Better still, Phil and PhishGuru are now being bundled into training programs for firms and organizations. Some organizations like to warn their members about phishing by the use of simulated phishing scams. (In one case, the U.S. Military Academy sent cadets an email signed by a fictitious colonel, asking for sensitive information. About 80 percent dutifully took the bait.) So Cranor and other CUPS faculty through a spinout company called Wombat Security Technologies are offering an added wrinkle: theyll write the simulated email, and rig it so that if you click on the baited link, you get an instant lesson from PhishGuru. Its taking advantage of a teachable moment, Cranor says. People are more likely to be receptive to teaching when they realize they just fell for an attack. Other projects now in the works at CUPS have to do with usable design. A survey of location-sharing services, with which you can use your GPS cell phone or Wi-Fi laptop to let people know where you are, found that systems on the market tend to lack good privacy-preference settings. Some leave you open to anyone who comes looking while others are confusing or dont give meaningful control. CUPS is piloting a system called Locaccino which, Cranor says, makes it easy to set up privacy rules. For instance, my students can access my location only while Im on campus on weekdays, but close friends and family any time, anywhere. (This too looks to be the basis of a spinout company.) And speaking of confusion: most public websites have privacy policies but its often hard to find them or figure out what they really mean. Thus Cranor and the CUPS team hope to create the equivalent of a nutrition label for privacy. Just as the labels on food products list the key ingredients in a standard format to help you comparison shop, the envisioned system would link with search engines to display various websites privacy policies in a uniform fashion. The CUPS team operates a search engine called Privacy Finder, which demonstrates the privacy nutrition label concept. The unifying theme in all CUPS projects is enabling people to make more informed choices more easily. This has been done in other areas; Cranor sees no reason it cant be done in online privacy and security. Weve already had faculty and students from many disciplines doing research, she says, and the students in the new Ph.D. program are going to help us develop new methodologies. Visit the CUPS website for more information about the CUPS Ph.D. program and CUPS research projects and to try out Anti-Phishing Phil, Locaccino and Privacy Finder, http://cups.cs.cmu.edu.

13 13

Tech Today Tech Today

T e c h

T o d ay

neW ReseARcH WIll HelP cOnsUMeRs TRAck HOMe eneRGy cOnsUMPTIOn

By Sherry STokeS

Most people dont know how to reduce energy costs in their homes. Sure, you can replace traditional light bulbs with energy-efficient ones or adjust the thermostat a few degrees, but it is difficult to quantify how these changes will reduce your power bill. However, in the not-too-distant future this may no longer be the case. A team of Carnegie Mellon researchers, lead by professors Lucio Soibelman, H. Scott Matthews and Jos M. F . Moura, received a $1.5 million grant from the National Science Foundation to track energy consumption in buildings. Soibelman says that their research relies on inexpensive sensor technology to track and quantify power consumption and it will ultimately teach people how to be more intelligent when they try to save energy. He compares energy consumption to grocery shopping. He explains that when you buy groceries, you get a receipt listing each item purchased. To save money the next time you shop, you can review the receipt and make decisions on how to cut expenses. Perhaps youll buy more chicken and less beef or substitute an off-brand cereal for a name brand. With electricity, we dont have that option. At the end of the month, the utility company sends a bill listing the total amount of power used and thats it. Consumers have no way of knowing how their energy habits or appliances impact their bills. Around four years ago, Soibelman, Matthews and Mario Berges, a Ph.D. student, set out to track energy consumption in buildings. Their first experiments took place in Porter Hall. They put sensors on all the buildings circuit breakers to collect data. The team realized this would be too expensive for home use. In a house, you have approximately 30 breakers. You would have to wire the sensors and do circuit tracing to determine what is feeding what, says Soibelman. It would cost approximately $10,000 to wire a typical house. If your power bill is $100 a month, it would be illogical to spend $10,000 to save $120 to $240 a year on electricity. Determined to find a cheaper way to track power usage, the team decided to apply non-intrusive load monitoring technology. The idea here is to have one sensor in the house that traces energy consumption. This sensor would monitor a matrix of data, i.e. watts, current, etc., as it passes through the wire and feed the data into complex computer algorithms for analysis. Soibelman explains that when you turn a light on or off, you get a bump in the graph or a transient that indicates a change from one state to another. We want to find specific characteristics in those transients that you could call signatures for each appliance, he says. Our long-term goal is to have a system in the house that could learn the patterns of how people use electricity and how to optimize electricity use.

14

engineering news

He envisions a system that relies on human-computer interaction. For example, a person would tell the power tracking system that they want to spend $100 a month for electricity. A couple of days later, the system would send back a message, predicting that the bill will be $150 if the consumer does not change his behavior. The system analyzes one or two days worth of energy usage in the home and then extrapolates data for the rest of the month. By understanding the homeowners behavior and having quantitative data on the amount of power each appliance uses, the system could then offer the consumer suggestions on how to cut costs. In 2007 the team consulted with researchers from Bosch Research and Technology Center North America, and together, with Bosch funding, they built the first prototype. The prototype proved feasible, and the team applied for a patent, but there were shortcomings. It takes a great deal of signal processing to accurately detect transients. We have 100% accuracy detecting heavy appliances and light loads. With medium usage, its hard to detect if a light is on in the bathroom or bedroom, says Soibelman. When Carnegie Mellon and Bosch applied for the NSF grant, CITs Jos Moura and Yuanwei Jin from University of Maryland Eastern Shore, both experts in signal processing, were added to the team. While the team is in place, Soibelman explains, It is not enough to have overwhelming amounts of data. We have to change how consumers use electricity. The role of human-computer interaction cannot be understated in this work, and that is why the Pittsburgh-based team is working with researchers from the Carnegie MellonPortugal Information and Communication Technologies Institute (ICTI), and in particular researchers from the University of Madeira. We are planning to install prototypes in 100 houses in Madeira, and the Portuguese team will track user behavior, says Soibelman. Mario Berges, who has worked on this project since its inception, explains, We know the technology, we know how to desegregate the data, and we even know what is consuming the energy in your house. But our Portuguese counterparts have the challenge of getting the data to the users and making users change their behavior. The Portuguese team will design an interface with the intention of changing user behavior for the long term. (This part of the research is not funded by the NSF . Instead, it is part of a project called SINAIS that is sponsored by the Portuguese National Science Foundation under the framework of the Carnegie Mellon Portugal Program.) While work is planned for Portugal, here in the Pittsburgh area, the sensor technology that CIT and Bosch developed is being tested for other purposes as well. In the city of McKeesport, Blueroof Technologies has built a number of high-tech cottages for elderly and disabled people. These homes are outfitted with advanced robotic and electronic devices, which have been designed by local universities and this includes CITs sensor technology. We have ideas that go beyond saving energy, says Soibelman. He and Berges explain that while it is not yet part of the plan, tracking systems could be developed to diagnose problems with appliances. Expanding on commercial applications, energy suppliers could benefit from tracking systems, as well. Surprisingly, utility companies dont have a lot of granular data on consumer behavior. These systems and the data they provide could help utility companies determine on- and offpeak rates, which could thwart blackouts. In turn, consumers could benefit because tracking systems would tell them the cheapest time of day to run major appliances. Financial incentives would certainly hasten the development of tracking technology, but understanding how we use energy and learning how to change our long-term behavior is critical if we are to someday achieve sustainable energy.

Our long-term goal is to have a system in the house that could learn the patterns of how people use electricity and how to optimize electricity use.

15 15

Tech Today