3 Questions: Community policing in the Global South

The concept of community policing gained wide acclaim in the U.S. when crime dropped drastically during the 1990s. In Chicago, Boston, and elsewhere, police departments established programs to build more local relationships, to better enhance community security. But how well does community policing work in other places? A new multicountry experiment co-led by MIT political scientist Fotini Christia found, perhaps surprisingly, that the policy had no impact in several countries across the Global South, from Africa to South America and Asia.

The results are detailed in a new edited volume, “Crime, Insecurity, and Community Policing: Experiments on Building Trust,” published this week by Cambridge University Press. The editors are Christia, the Ford International Professor of the Social Sciences in MIT’s Department of Political Science, director of the MIT Institute for Data, Systems, and Society, and director of the MIT Sociotechnical Systems Research Center; Graeme Blair of the University of California at Los Angeles; and Jeremy M. Weinstein of Stanford University. MIT News talked to Christia about the project.

Q: What is community policing, and how and where did you study it?

A: The general idea is that community policing, actually connecting the police and the community they are serving in direct ways, is very effective. Many of us have celebrated community policing, and we typically think of the 1990s Chicago and Boston experiences, where community policing was implemented and seen as wildly successful in reducing crime rates, gang violence, and homicide. This model has been broadly exported across the world, even though we don’t have much evidence that it works in contexts that have different resource capacities and institutional footprints.

Our study aims to understand if the hype around community policing is justified by measuring the effects of such policies globally, through field experiments, in six different settings in the Global South. In the same way that MIT’s J-PAL develops field experiments about an array of development interventions, we created programs, in cooperation with local governments, about policing. We studied if it works and how, across very diverse settings, including Uganda and Liberia in Africa, Colombia and Brazil in Latin America, and the Philippines and Pakistan in Asia.

The study, and book, is the result of collaborations with many police agencies. We also highlight how one can work with the police to understand and refine police practices and think very intentionally about all the ethical considerations around such collaborations. The researchers designed the interventions alongside six teams of academics who conducted the experiments, so the book also reflects an interesting experiment in how to put together a collaboration like this.

Q: What did you find?

A: What was fascinating was that we found that locally designed community policing interventions did not generate greater trust or cooperation between citizens and the police, and did not reduce crime in the six regions of the Global South where we carried out our research.

We looked at an array of different measures to evaluate the impact, such as changes in crime victimization, perceptions of police, as well as crime reporting, among others, and did not see any reductions in crime, whether measured in administrative data or in victimization surveys.

The null effects were not driven by concerns of police noncompliance with the intervention, crime displacement, or any heterogeneity in effects across sites, including individual experiences with the police.

Sometimes there is a bias against publishing so-called null results. But because we could show that it wasn’t due to methodological concerns, and because we were able to explain how such changes in resource-constrained environments would have to be preceded by structural reforms, the finding has been received as particularly compelling.

Q: Why did community policing not have an impact in these countries?

A: We felt that it was important to analyze why it doesn’t work. In the book, we highlight three challenges. One involves capacity issues: This is the developing world, and there are low-resource issues to begin with, in terms of the programs police can implement.

The second challenge is the principal-agent problem, the fact that the incentives of the police may not align in this case. For example, a station commander and supervisors may not appreciate the importance of adopting community policing, and line officers might not comply. Agency problems within the police are complex when it comes to mechanisms of accountability, and this may undermine the effectiveness of community policing.

A third challenge we highlight is the fact that, to the communities they serve, the police might not seem separate from the actual government. So, it may not be clear if police are seen as independent institutions acting in the best interest of the citizens.

We faced a lot of pushback when we were first presenting our results. The potential benefits of community policing is a story that resonates with many of us; it’s a narrative suggesting that connecting the police to a community has a significant and substantively positive effect. But the outcome didn’t come as a surprise to people from the Global South. They felt the lack of resources, and potential problems about autonomy and nonalignment, were real. 

de MIT News https://ift.tt/yAQiJ0D

From refugee to MIT graduate student

Mlen-Too Wesley has faded memories of his early childhood in Liberia, but the sharpest one has shaped his life.

Wesley was 4 years old when he and his family boarded a military airplane to flee the West African nation. At the time, the country was embroiled in a 14-year civil war that killed approximately 200,000 people, displaced about 750,000, and starved countless more. When Wesley’s grandmother told him he would enjoy a meal during his flight, Wesley knew his fortune had changed. Yet, his first instinct was to offer his food to the people he left behind.

“I made a decision right then to come back,” Wesley says. “Even as I grew older and spent more time in the United States, I knew I wanted to contribute to Liberia’s future.”

Today, the 38-year-old is committed to empowering Liberians through economic growth. Wesley looked to the MITx MicroMasters program in Data, Economics, and Design of Policy (DEDP) to achieve that goal. He examined issues such as micro-lending, state capture, and investment in health care in courses such as Foundations of Development Policy, Good Economics for Hard Times, and The Challenges of Global Poverty. Through case studies and research, Wesley discovered that economic incentives can encourage desired behaviors, curb corruption, and empower people.

“I couldn’t connect the dots”

Liberia is marred by corruption. According to Transparency International’s Corruptions Perception Index for 2023, Liberia scored 25 out of 100, with zero signifying the highest level of corruption. Yet, Wesley grew tired of textbooks and undergraduate professors saying that the status of Liberia and other African nations could be blamed entirely on corruption. Even worse, these sources gave Wesley the impression that nothing could be done to improve his native country. The sentiment frustrated him, he says.

“It struck me as flippant to attribute the challenges faced by billions of people to backward behaviors,” says Wesley. “There are several forces, internal and external, that have contributed to Liberia’s condition. If we really examine them, explore why things happened, and define the change we want, we can plot a way forward to a more prosperous future.”  

Driven to examine the economic, political, and social dynamics shaping his homeland and to fulfill his childhood promise, Wesley moved back to Africa in 2013. Over the next 10 years, he merged his interests in entrepreneurship, software development, and economics to better Liberia. He designed a forestry management platform that preserves Liberia’s natural resources, built an online queue for government hospitals to triage patients more effectively, and engineered data visualization tools to support renewable energy initiatives. Yet, to create the impact Wesley wanted, he needed to do more than collect data. He had to analyze and act on it in meaningful ways.

“I couldn’t connect the dots on why things are the way they are,” Wesley says.

“It wasn't just an academic experience for me”

Wesley knew he needed to dive deeper into data science, and looked to the MicroMasters in DEDP program to help him connect the dots. Established in 2017 by the Abdul Latif Jameel Poverty Action Lab (J-PAL) and MIT Open Learning, the MicroMasters in DEDP program is based on the Nobel Prize-winning work of MIT faculty members Esther Duflo, the Abdul Latif Jameel Professor of Poverty Alleviation and Development Economics, and Abhijit Banerjee, the Ford Foundation International Professor of Economics. Duflo and Banerjee’s research provided an entirely new approach to designing, implementing, and evaluating antipoverty initiatives throughout the world.

The MicroMasters in DEDP program provided the framework Wesley had sought nearly 20 years ago as an undergraduate student. He learned about novel economic incentives that stymied corruption and promoted education.

“It wasn't just an academic experience for me,” Wesley says. “The classes gave me the tools and the frameworks to analyze my own personal experiences.”

Wesley initially stumbled with the quantitative coursework. Having a demanding career, taking extension courses at another university, and being several years removed from college calculus courses took a toll on him. He had to retake some classes, especially Data Analysis for Social Scientists, several times before he could pass the proctored exam. His persistence paid off. Wesley earned his MicroMasters in DEDP credential in June 2023 and was also admitted into the MIT DEDP master’s program.

“The class twisted my brain in so many different ways,” Wesley says. “The fourth time taking Data Analysis, I began to understand it. I appreciate that MIT did not care that I did poorly on my first try. They cared that over time I understood the material.”

The program’s rigorous mathematics and statistics classes sparked in Wesley a passion for artificial intelligence, especially machine learning and natural language processing. Both provide more powerful ways to extract and interpret data, and Wesley has a special interest in mining qualitative sources for information. He plans to use these tools to compare national development plans over time and among different countries to determine if policymakers are recycling the same words and goals.

Once Wesley earns his master’s degree, he plans to return to Liberia and focus on international development. In the future, he hopes to lead a data-focused organization committed to improving the lives of people in Liberia and the United States.

“Thanks to MIT, I have the knowledge and tools to tackle real-world challenges that traditional economic models often overlook,” Wesley says.

de MIT News https://ift.tt/8IxcSMb

How mass migration remade postwar Europe

Migrants have become a flashpoint in global politics. But new research by an MIT political scientist, focused on West Germany and Poland after World War II, shows that in the long term, those countries developed stronger states, more prosperous economies, and more entrepreneurship after receiving a large influx of immigrants.

Those findings come from a close examination, at the local level over many decades, of the communities receiving migrants as millions of people relocated westward when Europe’s postwar borders were redrawn.

“I found that places experiencing large-scale displacement [immigration] wound up accumulating state capacity, versus places that did not,” says Volha Charnysh, the Ford Career Development Associate Professor in MIT’s Department of Political Science.

Charnysh’s new book, “Uprooted: How Post-WWII Population Transfers Remade Europe,” published by Cambridge University Press, challenges the notion that migrants have a negative impact on receiving communities.

The time frame of the analysis is important. Much discussion about refugees involves the short-term strains they place on institutions or the backlash they provoke in local communities. Charnysh’s research does reveal tensions in the postwar communities that received large numbers of refugees. But her work, distinctively, also quantifies long-run outcomes, producing a different overall picture.

As Charnysh writes in the book, “Counterintuitively, mass displacement ended up strengthening the state and improving economic performance in the long run.”

Extracting data from history

World War II wrought a colossal amount of death, destruction, and suffering, including the Holocaust, the genocide of about 6 million European Jews. The ensuing peace settlement among the Allied Powers led to large-scale population transfers. Poland saw its borders moved about 125 miles west; it was granted formerly German territory while ceding eastern territory to the Soviet Union. Its new region became 80 percent filled by new migrants, including Poles displaced from the east and voluntary migrants from other parts of the country and from abroad. West Germany received an influx of 12.5 million Germans displaced from Poland and other parts of Europe.

To study the impact of these population transfers, Charnysh used historical records to create four original quantitative datasets at the municipal and county level, while also examining archival documents, memoirs, and newspapers to better understand the texture of the time. The assignment of refugees to specific communities within Poland and West Germany amounted to a kind of historical natural experiment, allowing her to compare how the size and regional composition of the migrant population affected otherwise similar areas.

Additionally, studying forced displacement — as opposed to the movement of a self-selected group of immigrants — meant Charnysh could rigorously examine the scaled-up effects of mass migration.

“It has been an opportunity to study in a more robust way the consequences of displacement,” Charnysh says.

The Holocaust, followed by the redrawing of borders, expulsions, and mass relocations, appeared to increase the homogeneity of the populations within them: In 1931 Poland consisted of about one-third ethnic minorities, whereas after the war it became almost ethnically uniform. But one insight of Charnysh’s research is that shared ethnic or national identification does not guarantee social acceptance for migrants.

“Even if you just rearrange ethnically homogenous populations, new cleavages emerge,” Charnysh says. “People will not necessarily see others as being the same. Those who are displaced have suffered together, have a particular status in their new place, and realize their commonalities. For the native population, migrants’ arrival increased competition for jobs, housing, and state resources, so shared identities likewise emerged, and this ethnic homogeneity didn’t automatically translate into more harmonious relations.”

Yet, West Germany and Poland did assimilate these groups of immgrants into their countries. In both places, state capacity grew in the decades after the war, with the countries becoming better able to administer resources for their populations.

“The very problem, that migration and diversity can create conflict, can also create the demand for more state presence and, in cases where states are willing and able to step in, allow for the accumulation of greater state capacity over time,” Charnysh says.

State investment in migrant-receiving localities paid off. By the 1980s in West Germany, areas with greater postwar migration had higher levels of education, with more business enterprises being founded. That economic pattern emerged in Poland after it switched to a market economy in the 1990s.

Needed: Property rights and liberties

In “Uprooted,” Charnysh also discusses the conditions in which the example of West Germany and Poland may apply to other countries. For one thing, the phenomenon of migrants bolstering the economy is likeliest to occur where states offer what the scholars Daron Acemoglu and Simon Johnson of MIT and James Robinson of the University of Chicago have called “inclusive institutions,” such as property rights, additional liberties, and a commitment to the rule of law. Poland, while increasing its state capacity during the Cold War, did not realize the economic benefits of migration until the Cold War ended and it changed to a more democratic government.

Additionally, Charnysh observes, West Germany and Poland were granting citizenship to the migrants they received, making it easier for those migrants to assimilate and make demands on the state. “My complete account probably applies best to cases where migrants receive full citizenship rights,” she acknowledges.

“Uprooted” has earned praise from leading scholars. David Stasavage, dean for the social sciences and a professor of politics at New York University, has called the book a “pathbreaking study” that “upends what we thought we knew about the interaction between social cohesion and state capacity.” Charnysh’s research, he adds, “shows convincingly that areas with more diverse populations after the transfers saw greater improvements in state capacity and economic performance. This is a major addition to scholarship.”

Today there may be about 100 million displaced people around the world, including perhaps 14 million Ukrainians uprooted by war. Absorbing refugees may always be a matter of political contention. But as “Uprooted” shows, countries may realize benefits from it if they take a long-term perspective.

“When states treat refugees as temporary, they don’t provide opportunities for them to contribute and assimilate,” Charnysh says. “It’s not that I don’t think cultural differences matter to people, but it’s not as big a factor as state policies.” 

de MIT News https://ift.tt/b8f7kTB

An inflatable gastric balloon could help people lose weight

Gastric balloons — silicone balloons filled with air or saline and placed in the stomach — can help people lose weight by making them feel too full to overeat. However, this effect eventually can wear off as the stomach becomes used to the sensation of fullness.

To overcome that limitation, MIT engineers have designed a new type of gastric balloon that can be inflated and deflated as needed. In an animal study, they showed that inflating the balloon before a meal caused the animals to reduce their food intake by 60 percent.

This type of intervention could offer an alternative for people who don’t want to undergo more invasive treatments such as gastric bypass surgery, or people who don’t respond well to weight-loss drugs, the researchers say.

“The basic concept is we can have this balloon that is dynamic, so it would be inflated right before a meal and then you wouldn’t feel hungry. Then it would be deflated in between meals,” says Giovanni Traverso, an associate professor of mechanical engineering at MIT, a gastroenterologist at Brigham and Women’s Hospital, and the senior author of the study.

Neil Zixun Jia, who received a PhD from MIT in 2023, is the lead author of the paper, which appears today in the journal Device.

An inflatable balloon

Gastric balloons filled with saline are currently approved for use in the United States. These balloons stimulate a sense of fullness in the stomach, and studies have shown that they work well, but the benefits are often temporary.

“Gastric balloons do work initially. Historically, what has been seen is that the balloon is associated with weight loss. But then in general, the weight gain resumes the same trajectory,” Traverso says. “What we reasoned was perhaps if we had a system that simulates that fullness in a transient way, meaning right before a meal, that could be a way of inducing weight loss.”

To achieve a longer-lasting effect in patients, the researchers set out to design a device that could expand and contract on demand. They created two prototypes: One is a traditional balloon that inflates and deflates, and the other is a mechanical device with four arms that expand outward, pushing out an elastic polymer shell that presses on the stomach wall.

In animal tests, the researchers found that the mechanical-arm device could effectively expand to fill the stomach, but they ended up deciding to pursue the balloon option instead.

“Our sense was that the balloon probably distributed the force better, and down the line, if you have balloon that is applying the pressure, that is probably a safer approach in the long run,” Traverso says.

The researchers’ new balloon is similar to a traditional gastric balloon, but it is inserted into the stomach through an incision in the abdominal wall. The balloon is connected to an external controller that can be attached to the skin and contains a pump that inflates and deflates the balloon when needed. Inserting this device would be similar to the procedure used to place a feeding tube into a patient’s stomach, which is commonly done for people who are unable to eat or drink.

“If people, for example, are unable to swallow, they receive food through a tube like this. We know that we can keep tubes in for years, so there is already precedent for other systems that can stay in the body for a very long time. That gives us some confidence in the longer-term compatibility of this system,” Traverso says.

Reduced food intake

In tests in animals, the researchers found that inflating the balloon before meals led to a 60 percent reduction in the amount of food consumed. These studies were done over the course of a month, but the researchers now plan to do longer-term studies to see if this reduction leads to weight loss.

“The deployment for traditional gastric balloons is usually six months, if not more, and only then you will see good amount of weight loss. We will have to evaluate our device in a similar or longer time span to prove it really works better,” Jia says.

If developed for use in humans, the new gastric balloon could offer an alternative to existing obesity treatments. Other treatments for obesity include gastric bypass surgery, “stomach stapling” (a surgical procedure in which the stomach capacity is reduced), and drugs including GLP-1 receptor agonists such as semaglutide.

The gastric balloon could be an option for patients who are not good candidates for surgery or don’t respond well to weight-loss drugs, Traverso says.

“For certain patients who are higher-risk, who cannot undergo surgery, or did not tolerate the medication or had some other contraindication, there are limited options,” he says. “Traditional gastric balloons are still being used, but they come with a caveat that eventually the weight loss can plateau, so this is a way of trying to address that fundamental limitation.”

The research was funded by MIT’s Department of Mechanical Engineering, the Karl van Tassel Career Development Professorship, the Whitaker Health Sciences Fund Fellowship, the T.S. Lin Fellowship, the MIT Undergraduate Research Opportunities Program, and the Boston University Yawkey Funded Internship Program. 

de MIT News https://ift.tt/8ysK5Vj

Q&A: Transforming research through global collaborations

The MIT Global Seed Funds (GSF) program fosters global research collaborations with MIT faculty and their peers abroad — creating partnerships that tackle complex global issues, from climate change to health-care challenges and beyond. Administered by the MIT Center for International Studies (CIS), the GSF program has awarded more than $26 million to over 1,200 faculty research projects since its inception in 2008. Through its unique funding structure — comprising a general fund for unrestricted geographical use and several specific funds within individual countries, regions, and universities — GSF supports a wide range of projects. The current call for proposals from MIT faculty and researchers with principal investigator status is open until Dec. 10. 

CIS recently sat down with faculty recipients Josephine Carstensen and David McGee to discuss the value and impact GSF added to their research. Carstensen, the Gilbert W. Winslow Career Development Associate Professor of Civil and Environmental Engineering, generates computational designs for large-scale structures with the intent of designing novel low-carbon solutions. McGee, the William R. Kenan, Jr. Professor in the Department of Earth, Atmospheric and Planetary Sciences (EAPS), reconstructs the patterns, pace, and magnitudes of past hydro-climate changes.

Q: How did the Global Seed Funds program connect you with global partnerships related to your research?

Carstensen: One of the projects my lab is working on is to unlock the potential of complex cast-glass structures. Through our GSF partnership with researchers at TUDelft (Netherlands), my group was able to leverage our expertise in generative design algorithms alongside the TUDelft team, who are experts in the physical casting and fabrication of glass structures. Our initial connection to TUDelft was actually through one of my graduate students who was at a conference and met TUDelft researchers. He was inspired by their work and felt there could be synergy between our labs. The question then became: How do we connect with TUDelft? And that was what led us to the Global Seed Funds program. 

McGee: Our research is based in fieldwork conducted in partnership with experts who have a rich understanding of local environments. These locations range from lake basins in Chile and Argentina to caves in northern Mexico, Vietnam, and Madagascar. GSF has been invaluable for helping foster partnerships with collaborators and universities in these different locations, enabling the pilot work and relationship-building necessary to establish longer-term, externally funded projects.

Q: Tell us more about your GSF-funded work.

Carstensen: In my research group at MIT, we live mainly in a computational regime, and we do very little proof-of-concept testing. To that point, we do not even have the facilities nor experience to physically build large-scale structures, or even specialized structures. GSF has enabled us to connect with the researchers at TUDelft who do much more experimental testing than we do. Being able to work with the experts at TUDelft within their physical realm provided valuable insights into their way of approaching problems. And, likewise, the researchers at TUDelft benefited from our expertise. It has been fruitful in ways we couldn’t have imagined within our lab at MIT.

McGee: The collaborative work supported by the GSF has focused on reconstructing how past climate changes impacted rainfall patterns around the world, using natural archives like lake sediments and cave formations. One particularly successful project has been our work in caves in northeastern Mexico, which has been conducted in partnership with researchers from the National Autonomous University of Mexico (UNAM) and a local caving group. This project has involved several MIT undergraduate and graduate students, sponsored a research symposium in Mexico City, and helped us obtain funding from the National Science Foundation for a longer-term project.

Q: You both mentioned the involvement of your graduate students. How exactly has the GSF augmented the research experience of your students?

Carstensen: The collaboration has especially benefited the graduate students from both the MIT and TUDelft teams. The opportunity presented through this project to engage in research at an international peer institution has been extremely beneficial for their academic growth and maturity. It has facilitated training in new and complementary technical areas that they would not have had otherwise and allowed them to engage with leading world experts. An example of this aspect of the project's success is that the collaboration has inspired one of my graduate students to actively pursue postdoc opportunities in Europe (including at TU Delft) after his graduation.

McGee: MIT students have traveled to caves in northeastern Mexico and to lake basins in northern Chile to conduct fieldwork and build connections with local collaborators. Samples enabled by GSF-supported projects became the focus of two graduate students’ PhD theses, two EAPS undergraduate senior theses, and multiple UROP [Undergraduate Research Opportunity Program] projects.

Q: Were there any unexpected benefits to the work funded by GSF?

Carstensen: The success of this project would not have been possible without this specific international collaboration. Both the Delft and MIT teams bring highly different essential expertise that has been necessary for the successful project outcome. It allowed both the Delft and MIT teams to gain an in-depth understanding of the expertise areas and resources of the other collaborators. Both teams have been deeply inspired. This partnership has fueled conversations about potential future projects and provided multiple outcomes, including a plan to publish two journal papers on the project outcome. The first invited publication is being finalized now.

McGee: GSF’s focus on reciprocal exchange has enabled external collaborators to spend time at MIT, sharing their work and exchanging ideas. Other funding is often focused on sending MIT researchers and students out, but GSF has helped us bring collaborators here, making the relationship more equal. A GSF-supported visit by Argentinian researchers last year made it possible for them to interact not just with my group, but with students and faculty across EAPS.

de MIT News https://ift.tt/1uPyw8S

Photonic processor could enable ultrafast AI computations with extreme energy efficiency

The deep neural network models that power today’s most demanding machine-learning applications have grown so large and complex that they are pushing the limits of traditional electronic computing hardware.

Photonic hardware, which can perform machine-learning computations with light, offers a faster and more energy-efficient alternative. However, there are some types of neural network computations that a photonic device can’t perform, requiring the use of off-chip electronics or other techniques that hamper speed and efficiency.

Building on a decade of research, scientists from MIT and elsewhere have developed a new photonic chip that overcomes these roadblocks. They demonstrated a fully integrated photonic processor that can perform all the key computations of a deep neural network optically on the chip.

The optical device was able to complete the key computations for a machine-learning classification task in less than half a nanosecond while achieving more than 92 percent accuracy — performance that is on par with traditional hardware.

The chip, composed of interconnected modules that form an optical neural network, is fabricated using commercial foundry processes, which could enable the scaling of the technology and its integration into electronics.

In the long run, the photonic processor could lead to faster and more energy-efficient deep learning for computationally demanding applications like lidar, scientific research in astronomy and particle physics, or high-speed telecommunications.

“There are a lot of cases where how well the model performs isn’t the only thing that matters, but also how fast you can get an answer. Now that we have an end-to-end system that can run a neural network in optics, at a nanosecond time scale, we can start thinking at a higher level about applications and algorithms,” says Saumil Bandyopadhyay ’17, MEng ’18, PhD ’23, a visiting scientist in the Quantum Photonics and AI Group within the Research Laboratory of Electronics (RLE) and a postdoc at NTT Research, Inc., who is the lead author of a paper on the new chip.

Bandyopadhyay is joined on the paper by Alexander Sludds ’18, MEng ’19, PhD ’23; Nicholas Harris PhD ’17; Darius Bunandar PhD ’19; Stefan Krastanov, a former RLE research scientist who is now an assistant professor at the University of Massachusetts at Amherst; Ryan Hamerly, a visiting scientist at RLE and senior scientist at NTT Research; Matthew Streshinsky, a former silicon photonics lead at Nokia who is now co-founder and CEO of Enosemi; Michael Hochberg, president of Periplous, LLC; and Dirk Englund, a professor in the Department of Electrical Engineering and Computer Science, principal investigator of the Quantum Photonics and Artificial Intelligence Group and of RLE, and senior author of the paper. The research appears today in Nature Photonics.

Machine learning with light

Deep neural networks are composed of many interconnected layers of nodes, or neurons, that operate on input data to produce an output. One key operation in a deep neural network involves the use of linear algebra to perform matrix multiplication, which transforms data as it is passed from layer to layer.

But in addition to these linear operations, deep neural networks perform nonlinear operations that help the model learn more intricate patterns. Nonlinear operations, like activation functions, give deep neural networks the power to solve complex problems.

In 2017, Englund’s group, along with researchers in the lab of Marin Soljačić, the Cecil and Ida Green Professor of Physics, demonstrated an optical neural network on a single photonic chip that could perform matrix multiplication with light.

But at the time, the device couldn’t perform nonlinear operations on the chip. Optical data had to be converted into electrical signals and sent to a digital processor to perform nonlinear operations.

“Nonlinearity in optics is quite challenging because photons don’t interact with each other very easily. That makes it very power consuming to trigger optical nonlinearities, so it becomes challenging to build a system that can do it in a scalable way,” Bandyopadhyay explains.

They overcame that challenge by designing devices called nonlinear optical function units (NOFUs), which combine electronics and optics to implement nonlinear operations on the chip.

The researchers built an optical deep neural network on a photonic chip using three layers of devices that perform linear and nonlinear operations.

A fully-integrated network

At the outset, their system encodes the parameters of a deep neural network into light. Then, an array of programmable beamsplitters, which was demonstrated in the 2017 paper, performs matrix multiplication on those inputs.

The data then pass to programmable NOFUs, which implement nonlinear functions by siphoning off a small amount of light to photodiodes that convert optical signals to electric current. This process, which eliminates the need for an external amplifier, consumes very little energy.

“We stay in the optical domain the whole time, until the end when we want to read out the answer. This enables us to achieve ultra-low latency,” Bandyopadhyay says.

Achieving such low latency enabled them to efficiently train a deep neural network on the chip, a process known as in situ training that typically consumes a huge amount of energy in digital hardware.

“This is especially useful for systems where you are doing in-domain processing of optical signals, like navigation or telecommunications, but also in systems that you want to learn in real time,” he says.

The photonic system achieved more than 96 percent accuracy during training tests and more than 92 percent accuracy during inference, which is comparable to traditional hardware. In addition, the chip performs key computations in less than half a nanosecond.     

“This work demonstrates that computing — at its essence, the mapping of inputs to outputs — can be compiled onto new architectures of linear and nonlinear physics that enable a fundamentally different scaling law of computation versus effort needed,” says Englund.

The entire circuit was fabricated using the same infrastructure and foundry processes that produce CMOS computer chips. This could enable the chip to be manufactured at scale, using tried-and-true techniques that introduce very little error into the fabrication process.

Scaling up their device and integrating it with real-world electronics like cameras or telecommunications systems will be a major focus of future work, Bandyopadhyay says. In addition, the researchers want to explore algorithms that can leverage the advantages of optics to train systems faster and with better energy efficiency.

This research was funded, in part, by the U.S. National Science Foundation, the U.S. Air Force Office of Scientific Research, and NTT Research.

de MIT News https://ift.tt/EIPmink