Research

Subterms

    Latest story

    175 Shares51 Views

    This tiny chip can safeguard user data while enabling efficient computing on a smartphone

    by

    Health-monitoring apps can help people manage chronic diseases or stay on track with fitness goals, using nothing more than a smartphone. However, these apps can be slow and energy-inefficient because the vast machine-learning models that power them must be shuttled between a smartphone and a central memory server.

    Engineers often speed things up using hardware that reduces the need to move so much data back and forth. While these machine-learning accelerators can streamline computation, they are susceptible to attackers who can steal secret information.

    To reduce this vulnerability, researchers from MIT and the MIT-IBM Watson AI Lab created a machine-learning accelerator that is resistant to the two most common types of attacks. Their chip can keep a user’s health records, financial information, or other sensitive data private while still enabling huge AI models to run efficiently on devices.

    The team developed several optimizations that enable strong security while only slightly slowing the device. Moreover, the added security does not impact the accuracy of computations. This machine-learning accelerator could be particularly beneficial for demanding AI applications like augmented and virtual reality or autonomous driving.

    While implementing the chip would make a device slightly more expensive and less energy-efficient, that is sometimes a worthwhile price to pay for security, says lead author Maitreyi Ashok, an electrical engineering and computer science (EECS) graduate student at MIT.

    “It is important to design with security in mind from the ground up. If you are trying to add even a minimal amount of security after a system has been designed, it is prohibitively expensive. We were able to effectively balance a lot of these tradeoffs during the design phase,” says Ashok.

    Her co-authors include Saurav Maji, an EECS graduate student; Xin Zhang and John Cohn of the MIT-IBM Watson AI Lab; and senior author Anantha Chandrakasan, MIT’s chief innovation and strategy officer, dean of the School of Engineering, and the Vannevar Bush Professor of EECS. The research will be presented at the IEEE Custom Integrated Circuits Conference.

    Side-channel susceptibility

    The researchers targeted a type of machine-learning accelerator called digital in-memory compute. A digital IMC chip performs computations inside a device’s memory, where pieces of a machine-learning model are stored after being moved over from a central server.

    The entire model is too big to store on the device, but by breaking it into pieces and reusing those pieces as much as possible, IMC chips reduce the amount of data that must be moved back and forth.

    But IMC chips can be susceptible to hackers. In a side-channel attack, a hacker monitors the chip’s power consumption and uses statistical techniques to reverse-engineer data as the chip computes. In a bus-probing attack, the hacker can steal bits of the model and dataset by probing the communication between the accelerator and the off-chip memory.

    Digital IMC speeds computation by performing millions of operations at once, but this complexity makes it tough to prevent attacks using traditional security measures, Ashok says.

    She and her collaborators took a three-pronged approach to blocking side-channel and bus-probing attacks.

    First, they employed a security measure where data in the IMC are split into random pieces. For instance, a bit zero might be split into three bits that still equal zero after a logical operation. The IMC never computes with all pieces in the same operation, so a side-channel attack could never reconstruct the real information.

    But for this technique to work, random bits must be added to split the data. Because digital IMC performs millions of operations at once, generating so many random bits would involve too much computing. For their chip, the researchers found a way to simplify computations, making it easier to effectively split data while eliminating the need for random bits.

    Second, they prevented bus-probing attacks using a lightweight cipher that encrypts the model stored in off-chip memory. This lightweight cipher only requires simple computations. In addition, they only decrypted the pieces of the model stored on the chip when necessary.

    Third, to improve security, they generated the key that decrypts the cipher directly on the chip, rather than moving it back and forth with the model. They generated this unique key from random variations in the chip that are introduced during manufacturing, using what is known as a physically unclonable function.

    “Maybe one wire is going to be a little bit thicker than another. We can use these variations to get zeros and ones out of a circuit. For every chip, we can get a random key that should be consistent because these random properties shouldn’t change significantly over time,” Ashok explains.

    They reused the memory cells on the chip, leveraging the imperfections in these cells to generate the key. This requires less computation than generating a key from scratch.

    “As security has become a critical issue in the design of edge devices, there is a need to develop a complete system stack focusing on secure operation. This work focuses on security for machine-learning workloads and describes a digital processor that uses cross-cutting optimization. It incorporates encrypted data access between memory and processor, approaches to preventing side-channel attacks using randomization, and exploiting variability to generate unique codes. Such designs are going to be critical in future mobile devices,” says Chandrakasan.

    Safety testing

    To test their chip, the researchers took on the role of hackers and tried to steal secret information using side-channel and bus-probing attacks.

    Even after making millions of attempts, they couldn’t reconstruct any real information or extract pieces of the model or dataset. The cipher also remained unbreakable. By contrast, it took only about 5,000 samples to steal information from an unprotected chip.

    The addition of security did reduce the energy efficiency of the accelerator, and it also required a larger chip area, which would make it more expensive to fabricate.

    The team is planning to explore methods that could reduce the energy consumption and size of their chip in the future, which would make it easier to implement at scale.

    “As it becomes too expensive, it becomes harder to convince someone that security is critical. Future work could explore these tradeoffs. Maybe we could make it a little less secure but easier to implement and less expensive,” Ashok says.

    The research is funded, in part, by the MIT-IBM Watson AI Lab, the National Science Foundation, and a Mathworks Engineering Fellowship. More

    More stories

    • 125 Shares60 Views

      in Data Management & Statistics

      Using deep learning to image the Earth’s planetary boundary layer

      by

      Although the troposphere is often thought of as the closest layer of the atmosphere to the Earth’s surface, the planetary boundary layer (PBL) — the lowest layer of the troposphere — is actually the part that most significantly influences weather near the surface. In the 2018 planetary science decadal survey, the PBL was raised as an important scientific issue that has the potential to enhance storm forecasting and improve climate projections.  

      “The PBL is where the surface interacts with the atmosphere, including exchanges of moisture and heat that help lead to severe weather and a changing climate,” says Adam Milstein, a technical staff member in Lincoln Laboratory’s Applied Space Systems Group. “The PBL is also where humans live, and the turbulent movement of aerosols throughout the PBL is important for air quality that influences human health.” 

      Although vital for studying weather and climate, important features of the PBL, such as its height, are difficult to resolve with current technology. In the past four years, Lincoln Laboratory staff have been studying the PBL, focusing on two different tasks: using machine learning to make 3D-scanned profiles of the atmosphere, and resolving the vertical structure of the atmosphere more clearly in order to better predict droughts.  

      This PBL-focused research effort builds on more than a decade of related work on fast, operational neural network algorithms developed by Lincoln Laboratory for NASA missions. These missions include the Time-Resolved Observations of Precipitation structure and storm Intensity with a Constellation of Smallsats (TROPICS) mission as well as Aqua, a satellite that collects data about Earth’s water cycle and observes variables such as ocean temperature, precipitation, and water vapor in the atmosphere. These algorithms retrieve temperature and humidity from the satellite instrument data and have been shown to significantly improve the accuracy and usable global coverage of the observations over previous approaches. For TROPICS, the algorithms help retrieve data that are used to characterize a storm’s rapidly evolving structures in near-real time, and for Aqua, it has helped increase forecasting models, drought monitoring, and fire prediction. 

      These operational algorithms for TROPICS and Aqua are based on classic “shallow” neural networks to maximize speed and simplicity, creating a one-dimensional vertical profile for each spectral measurement collected by the instrument over each location. While this approach has improved observations of the atmosphere down to the surface overall, including the PBL, laboratory staff determined that newer “deep” learning techniques that treat the atmosphere over a region of interest as a three-dimensional image are needed to improve PBL details further.

      “We hypothesized that deep learning and artificial intelligence (AI) techniques could improve on current approaches by incorporating a better statistical representation of 3D temperature and humidity imagery of the atmosphere into the solutions,” Milstein says. “But it took a while to figure out how to create the best dataset — a mix of real and simulated data; we needed to prepare to train these techniques.”

      The team collaborated with Joseph Santanello of the NASA Goddard Space Flight Center and William Blackwell, also of the Applied Space Systems Group, in a recent NASA-funded effort showing that these retrieval algorithms can improve PBL detail, including more accurate determination of the PBL height than the previous state of the art. 

      While improved knowledge of the PBL is broadly useful for increasing understanding of climate and weather, one key application is prediction of droughts. According to a Global Drought Snapshot report released last year, droughts are a pressing planetary issue that the global community needs to address. Lack of humidity near the surface, specifically at the level of the PBL, is the leading indicator of drought. While previous studies using remote-sensing techniques have examined the humidity of soil to determine drought risk, studying the atmosphere can help predict when droughts will happen.  

      In an effort funded by Lincoln Laboratory’s Climate Change Initiative, Milstein, along with laboratory staff member Michael Pieper, are working with scientists at NASA’s Jet Propulsion Laboratory (JPL) to use neural network techniques to improve drought prediction over the continental United States. While the work builds off of existing operational work JPL has done incorporating (in part) the laboratory’s operational “shallow” neural network approach for Aqua, the team believes that this work and the PBL-focused deep learning research work can be combined to further improve the accuracy of drought prediction. 

      “Lincoln Laboratory has been working with NASA for more than a decade on neural network algorithms for estimating temperature and humidity in the atmosphere from space-borne infrared and microwave instruments, including those on the Aqua spacecraft,” Milstein says. “Over that time, we have learned a lot about this problem by working with the science community, including learning about what scientific challenges remain. Our long experience working on this type of remote sensing with NASA scientists, as well as our experience with using neural network techniques, gave us a unique perspective.”

      According to Milstein, the next step for this project is to compare the deep learning results to datasets from the National Oceanic and Atmospheric Administration, NASA, and the Department of Energy collected directly in the PBL using radiosondes, a type of instrument flown on a weather balloon. “These direct measurements can be considered a kind of ‘ground truth’ to quantify the accuracy of the techniques we have developed,” Milstein says.

      This improved neural network approach holds promise to demonstrate drought prediction that can exceed the capabilities of existing indicators, Milstein says, and to be a tool that scientists can rely on for decades to come. More

    • 163 Shares170 Views

      in Data Management & Statistics

      This 3D printer can figure out how to print with an unknown material

      by

      While 3D printing has exploded in popularity, many of the plastic materials these printers use to create objects cannot be easily recycled. While new sustainable materials are emerging for use in 3D printing, they remain difficult to adopt because 3D printer settings need to be adjusted for each material, a process generally done by hand.

      To print a new material from scratch, one must typically set up to 100 parameters in software that controls how the printer will extrude the material as it fabricates an object. Commonly used materials, like mass-manufactured polymers, have established sets of parameters that were perfected through tedious, trial-and-error processes.

      But the properties of renewable and recyclable materials can fluctuate widely based on their composition, so fixed parameter sets are nearly impossible to create. In this case, users must come up with all these parameters by hand.

      Researchers tackled this problem by developing a 3D printer that can automatically identify the parameters of an unknown material on its own.

      A collaborative team from MIT’s Center for Bits and Atoms (CBA), the U.S. National Institute of Standards and Technology (NIST), and the National Center for Scientific Research in Greece (Demokritos) modified the extruder, the “heart” of a 3D printer, so it can measure the forces and flow of a material.

      These data, gathered through a 20-minute test, are fed into a mathematical function that is used to automatically generate printing parameters. These parameters can be entered into off-the-shelf 3D printing software and used to print with a never-before-seen material. 

      The automatically generated parameters can replace about half of the parameters that typically must be tuned by hand. In a series of test prints with unique materials, including several renewable materials, the researchers showed that their method can consistently produce viable parameters.

      This research could help to reduce the environmental impact of additive manufacturing, which typically relies on nonrecyclable polymers and resins derived from fossil fuels.

      “In this paper, we demonstrate a method that can take all these interesting materials that are bio-based and made from various sustainable sources and show that the printer can figure out by itself how to print those materials. The goal is to make 3D printing more sustainable,” says senior author Neil Gershenfeld, who leads CBA.

      His co-authors include first author Jake Read a graduate student in the CBA who led the printer development; Jonathan Seppala, a chemical engineer in the Materials Science and Engineering Division of NIST; Filippos Tourlomousis, a former CBA postdoc who now heads the Autonomous Science Lab at Demokritos; James Warren, who leads the Materials Genome Program at NIST; and Nicole Bakker, a research assistant at CBA. The research is published in the journal Integrating Materials and Manufacturing Innovation.

      Shifting material properties

      In fused filament fabrication (FFF), which is often used in rapid prototyping, molten polymers are extruded through a heated nozzle layer-by-layer to build a part. Software, called a slicer, provides instructions to the machine, but the slicer must be configured to work with a particular material.

      Using renewable or recycled materials in an FFF 3D printer is especially challenging because there are so many variables that affect the material properties.

      For instance, a bio-based polymer or resin might be composed of different mixes of plants based on the season. The properties of recycled materials also vary widely based on what is available to recycle.

      “In ‘Back to the Future,’ there is a ‘Mr. Fusion’ blender where Doc just throws whatever he has into the blender and it works [as a power source for the DeLorean time machine]. That is the same idea here. Ideally, with plastics recycling, you could just shred what you have and print with it. But, with current feed-forward systems, that won’t work because if your filament changes significantly during the print, everything would break,” Read says.

      To overcome these challenges, the researchers developed a 3D printer and workflow to automatically identify viable process parameters for any unknown material.

      They started with a 3D printer their lab had previously developed that can capture data and provide feedback as it operates. The researchers added three instruments to the machine’s extruder that take measurements which are used to calculate parameters.

      A load cell measures the pressure being exerted on the printing filament, while a feed rate sensor measures the thickness of the filament and the actual rate at which it is being fed through the printer.

      “This fusion of measurement, modeling, and manufacturing is at the heart of the collaboration between NIST and CBA, as we work develop what we’ve termed ‘computational metrology,’” says Warren.

      These measurements can be used to calculate the two most important, yet difficult to determine, printing parameters: flow rate and temperature. Nearly half of all print settings in standard software are related to these two parameters. 

      Deriving a dataset

      Once they had the new instruments in place, the researchers developed a 20-minute test that generates a series of temperature and pressure readings at different flow rates. Essentially, the test involves setting the print nozzle at its hottest temperature, flowing the material through at a fixed rate, and then turning the heater off.

      “It was really difficult to figure out how to make that test work. Trying to find the limits of the extruder means that you are going to break the extruder pretty often while you are testing it. The notion of turning the heater off and just passively taking measurements was the ‘aha’ moment,” says Read.

      These data are entered into a function that automatically generates real parameters for the material and machine configuration, based on relative temperature and pressure inputs. The user can then enter those parameters into 3D printing software and generate instructions for the printer.

      In experiments with six different materials, several of which were bio-based, the method automatically generated viable parameters that consistently led to successful prints of a complex object.

      Moving forward, the researchers plan to integrate this process with 3D printing software so parameters don’t need to be entered manually. In addition, they want to enhance their workflow by incorporating a thermodynamic model of the hot end, which is the part of the printer that melts the filament.

      This collaboration is now more broadly developing computational metrology, in which the output of a measurement is a predictive model rather than just a parameter. The researchers will be applying this in other areas of advanced manufacturing, as well as in expanding access to metrology.

      “By developing a new method for the automatic generation of process parameters for fused filament fabrication, this study opens the door to the use of recycled and bio-based filaments that have variable and unknown behaviors. Importantly, this enhances the potential for digital manufacturing technology to utilize locally sourced sustainable materials,” says Alysia Garmulewicz, an associate professor in the Faculty of Administration and Economics at the University of Santiago in Chile who was not involved with this work.

      This research is supported, in part, by the National Institute of Standards and Technology and the Center for Bits and Atoms Consortia. More

    • 38 Shares125 Views

      in Data Management & Statistics

      Characterizing social networks

      by

      People tend to connect with others who are like them. Alumni from the same alma mater are more likely to collaborate over a research project together, or individuals with the same political beliefs are more likely to join the same political parties, attend rallies, and engage in online discussions. This sociology concept, called homophily, has been observed in many network science studies. But if like-minded individuals cluster in online and offline spaces to reinforce each other’s ideas and form synergies, what does that mean for society?

      Researchers at MIT wanted to investigate homophily further to understand how groups of three or more interact in complex societal settings. Prior research on understanding homophily has studied relationships between pairs of people. For example, when two members of Congress co-sponsor a bill, they are likely to be from the same political party.

      However, less is known about whether group interactions between three or more people are likely to occur between similar individuals. If three members of Congress co-sponsor a bill together, are all three likely to be members of the same party, or would we expect more bipartisanship? When the researchers tried to extend traditional methods to measure homophily in these larger group interactions, they found the results can be misleading.

      “We found that homophily observed in pairs, or one-to-one interactions, can make it seem like there’s more homophily in larger groups than there really is,” says Arnab Sarker, graduate student in the Institute for Data, Systems and Society (IDSS) and lead author of the study published in Proceedings of the National Academy of Sciences. “The previous measure didn’t account for the way in which two people already know each other in friendship settings,” he adds.

      To address this issue, Sarker, along with co-authors Natalie Northrup ’22 and Ali Jadbabaie, the JR East Professor of Engineering, head of the Department of Civil and Environmental Engineering, and core faculty member of IDSS, developed a new way of measuring homophily. Borrowing tools from algebraic topology, a subfield in mathematics typically applied in physics, they developed a new measure to understand whether homophily occurred in group interactions.

      The new measure, called simplicial homophily, separates the homophily seen in one-on-one interactions from those in larger group interactions and is based on the mathematical concept of a simplicial complex. The researchers tested this new measure with real-world data from 16 different datasets and found that simplicial homophily provides more accurate insights into how similar things interact in larger groups. Interestingly, the new measure can better identify instances where there is a lack of similarity in larger group interactions, thus rectifying a weakness observed in the previous measure.

      One such example of this instance was demonstrated in the dataset from the global hotel booking website, Trivago. They found that when travelers are looking at two hotels in one session, they often pick hotels that are close to one another geographically. But when they look at more than two hotels in one session, they are more likely to be searching for hotels that are farther apart from one another (for example, if they are taking a vacation with multiple stops). The new method showed “anti-homophily” — instead of similar hotels being chosen together, different hotels were chosen together.

      “Our measure controls for pairwise connections and is suggesting that there’s more diversity in the hotels that people are looking for as group size increases, which is an interesting economic result,” says Sarker.

      Additionally, they discovered that simplicial homophily can help identify when certain characteristics are important for predicting if groups will interact in the future. They found that when there’s a lot of similarity or a lot of difference between individuals who already interact in groups, then knowing individual characteristics can help predict their connection to each other in the future.

      Northrup was an undergraduate researcher on the project and worked with Sarker and Jadbabaie over three semesters before she graduated. The project gave her an opportunity to take some of the concepts she learned in the classroom and apply them.

      “Working on this project, I really dove into building out the higher-order network model, and understanding the network, the math, and being able to implement it at a large scale,” says Northrup, who was in the civil and environmental engineering systems track with a double major in economics.

      The new measure opens up opportunities to study complex group interactions in a broad range of network applications, from ecology to traffic and socioeconomics. One of the areas Sarker has interest in exploring is the group dynamics of people finding jobs through social networks. “Does higher-order homophily affect how people get information about jobs?” he asks.    

      Northrup adds that it could also be used to evaluate interventions or specific policies to connect people with job opportunities outside of their network. “You can even use it as a measurement to evaluate how effective that might be.”

      The research was supported through funding from a Vannevar Bush Fellowship from the Office of the U.S. Secretary of Defense and from the U.S. Army Research Office Multidisciplinary University Research Initiative. More

    • 150 Shares189 Views

      in Data Management & Statistics

      New software enables blind and low-vision users to create interactive, accessible charts

      by

      A growing number of tools enable users to make online data representations, like charts, that are accessible for people who are blind or have low vision. However, most tools require an existing visual chart that can then be converted into an accessible format.

      This creates barriers that prevent blind and low-vision users from building their own custom data representations, and it can limit their ability to explore and analyze important information.

      A team of researchers from MIT and University College London (UCL) wants to change the way people think about accessible data representations.

      They created a software system called Umwelt (which means “environment” in German) that can enable blind and low-vision users to build customized, multimodal data representations without needing an initial visual chart.

      Umwelt, an authoring environment designed for screen-reader users, incorporates an editor that allows someone to upload a dataset and create a customized representation, such as a scatterplot, that can include three modalities: visualization, textual description, and sonification. Sonification involves converting data into nonspeech audio.

      The system, which can represent a variety of data types, includes a viewer that enables a blind or low-vision user to interactively explore a data representation, seamlessly switching between each modality to interact with data in a different way.

      The researchers conducted a study with five expert screen-reader users who found Umwelt to be useful and easy to learn. In addition to offering an interface that empowered them to create data representations — something they said was sorely lacking — the users said Umwelt could facilitate communication between people who rely on different senses.

      “We have to remember that blind and low-vision people aren’t isolated. They exist in these contexts where they want to talk to other people about data,” says Jonathan Zong, an electrical engineering and computer science (EECS) graduate student and lead author of a paper introducing Umwelt. “I am hopeful that Umwelt helps shift the way that researchers think about accessible data analysis. Enabling the full participation of blind and low-vision people in data analysis involves seeing visualization as just one piece of this bigger, multisensory puzzle.”

      Joining Zong on the paper are fellow EECS graduate students Isabella Pedraza Pineros and Mengzhu “Katie” Chen; Daniel Hajas, a UCL researcher who works with the Global Disability Innovation Hub; and senior author Arvind Satyanarayan, associate professor of computer science at MIT who leads the Visualization Group in the Computer Science and Artificial Intelligence Laboratory. The paper will be presented at the ACM Conference on Human Factors in Computing.

      De-centering visualization

      The researchers previously developed interactive interfaces that provide a richer experience for screen reader users as they explore accessible data representations. Through that work, they realized most tools for creating such representations involve converting existing visual charts.

      Aiming to decenter visual representations in data analysis, Zong and Hajas, who lost his sight at age 16, began co-designing Umwelt more than a year ago.

      At the outset, they realized they would need to rethink how to represent the same data using visual, auditory, and textual forms.

      “We had to put a common denominator behind the three modalities. By creating this new language for representations, and making the output and input accessible, the whole is greater than the sum of its parts,” says Hajas.

      To build Umwelt, they first considered what is unique about the way people use each sense.

      For instance, a sighted user can see the overall pattern of a scatterplot and, at the same time, move their eyes to focus on different data points. But for someone listening to a sonification, the experience is linear since data are converted into tones that must be played back one at a time.

      “If you are only thinking about directly translating visual features into nonvisual features, then you miss out on the unique strengths and weaknesses of each modality,” Zong adds.

      They designed Umwelt to offer flexibility, enabling a user to switch between modalities easily when one would better suit their task at a given time.

      To use the editor, one uploads a dataset to Umwelt, which employs heuristics to automatically creates default representations in each modality.

      If the dataset contains stock prices for companies, Umwelt might generate a multiseries line chart, a textual structure that groups data by ticker symbol and date, and a sonification that uses tone length to represent the price for each date, arranged by ticker symbol.

      The default heuristics are intended to help the user get started.

      “In any kind of creative tool, you have a blank-slate effect where it is hard to know how to begin. That is compounded in a multimodal tool because you have to specify things in three different representations,” Zong says.

      The editor links interactions across modalities, so if a user changes the textual description, that information is adjusted in the corresponding sonification. Someone could utilize the editor to build a multimodal representation, switch to the viewer for an initial exploration, then return to the editor to make adjustments.

      Helping users communicate about data

      To test Umwelt, they created a diverse set of multimodal representations, from scatterplots to multiview charts, to ensure the system could effectively represent different data types. Then they put the tool in the hands of five expert screen reader users.

      Study participants mostly found Umwelt to be useful for creating, exploring, and discussing data representations. One user said Umwelt was like an “enabler” that decreased the time it took them to analyze data. The users agreed that Umwelt could help them communicate about data more easily with sighted colleagues.

      “What stands out about Umwelt is its core philosophy of de-emphasizing the visual in favor of a balanced, multisensory data experience. Often, nonvisual data representations are relegated to the status of secondary considerations, mere add-ons to their visual counterparts. However, visualization is merely one aspect of data representation. I appreciate their efforts in shifting this perception and embracing a more inclusive approach to data science,” says JooYoung Seo, an assistant professor in the School of Information Sciences at the University of Illinois at Urbana-Champagne, who was not involved with this work.

      Moving forward, the researchers plan to create an open-source version of Umwelt that others can build upon. They also want to integrate tactile sensing into the software system as an additional modality, enabling the use of tools like refreshable tactile graphics displays.

      “In addition to its impact on end users, I am hoping that Umwelt can be a platform for asking scientific questions around how people use and perceive multimodal representations, and how we can improve the design beyond this initial step,” says Zong.

      This work was supported, in part, by the National Science Foundation and the MIT Morningside Academy for Design Fellowship. More

    • 138 Shares139 Views

      in Data Management & Statistics

      AI generates high-quality images 30 times faster in a single step

      by

      In our current age of artificial intelligence, computers can generate their own “art” by way of diffusion models, iteratively adding structure to a noisy initial state until a clear image or video emerges. Diffusion models have suddenly grabbed a seat at everyone’s table: Enter a few words and experience instantaneous, dopamine-spiking dreamscapes at the intersection of reality and fantasy. Behind the scenes, it involves a complex, time-intensive process requiring numerous iterations for the algorithm to perfect the image.

      MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) researchers have introduced a new framework that simplifies the multi-step process of traditional diffusion models into a single step, addressing previous limitations. This is done through a type of teacher-student model: teaching a new computer model to mimic the behavior of more complicated, original models that generate images. The approach, known as distribution matching distillation (DMD), retains the quality of the generated images and allows for much faster generation. 

      “Our work is a novel method that accelerates current diffusion models such as Stable Diffusion and DALLE-3 by 30 times,” says Tianwei Yin, an MIT PhD student in electrical engineering and computer science, CSAIL affiliate, and the lead researcher on the DMD framework. “This advancement not only significantly reduces computational time but also retains, if not surpasses, the quality of the generated visual content. Theoretically, the approach marries the principles of generative adversarial networks (GANs) with those of diffusion models, achieving visual content generation in a single step — a stark contrast to the hundred steps of iterative refinement required by current diffusion models. It could potentially be a new generative modeling method that excels in speed and quality.”

      This single-step diffusion model could enhance design tools, enabling quicker content creation and potentially supporting advancements in drug discovery and 3D modeling, where promptness and efficacy are key.

      Distribution dreams

      DMD cleverly has two components. First, it uses a regression loss, which anchors the mapping to ensure a coarse organization of the space of images to make training more stable. Next, it uses a distribution matching loss, which ensures that the probability to generate a given image with the student model corresponds to its real-world occurrence frequency. To do this, it leverages two diffusion models that act as guides, helping the system understand the difference between real and generated images and making training the speedy one-step generator possible.

      The system achieves faster generation by training a new network to minimize the distribution divergence between its generated images and those from the training dataset used by traditional diffusion models. “Our key insight is to approximate gradients that guide the improvement of the new model using two diffusion models,” says Yin. “In this way, we distill the knowledge of the original, more complex model into the simpler, faster one, while bypassing the notorious instability and mode collapse issues in GANs.” 

      Yin and colleagues used pre-trained networks for the new student model, simplifying the process. By copying and fine-tuning parameters from the original models, the team achieved fast training convergence of the new model, which is capable of producing high-quality images with the same architectural foundation. “This enables combining with other system optimizations based on the original architecture to further accelerate the creation process,” adds Yin. 

      When put to the test against the usual methods, using a wide range of benchmarks, DMD showed consistent performance. On the popular benchmark of generating images based on specific classes on ImageNet, DMD is the first one-step diffusion technique that churns out pictures pretty much on par with those from the original, more complex models, rocking a super-close Fréchet inception distance (FID) score of just 0.3, which is impressive, since FID is all about judging the quality and diversity of generated images. Furthermore, DMD excels in industrial-scale text-to-image generation and achieves state-of-the-art one-step generation performance. There’s still a slight quality gap when tackling trickier text-to-image applications, suggesting there’s a bit of room for improvement down the line. 

      Additionally, the performance of the DMD-generated images is intrinsically linked to the capabilities of the teacher model used during the distillation process. In the current form, which uses Stable Diffusion v1.5 as the teacher model, the student inherits limitations such as rendering detailed depictions of text and small faces, suggesting that DMD-generated images could be further enhanced by more advanced teacher models. 

      “Decreasing the number of iterations has been the Holy Grail in diffusion models since their inception,” says Fredo Durand, MIT professor of electrical engineering and computer science, CSAIL principal investigator, and a lead author on the paper. “We are very excited to finally enable single-step image generation, which will dramatically reduce compute costs and accelerate the process.” 

      “Finally, a paper that successfully combines the versatility and high visual quality of diffusion models with the real-time performance of GANs,” says Alexei Efros, a professor of electrical engineering and computer science at the University of California at Berkeley who was not involved in this study. “I expect this work to open up fantastic possibilities for high-quality real-time visual editing.” 

      Yin and Durand’s fellow authors are MIT electrical engineering and computer science professor and CSAIL principal investigator William T. Freeman, as well as Adobe research scientists Michaël Gharbi SM ’15, PhD ’18; Richard Zhang; Eli Shechtman; and Taesung Park. Their work was supported, in part, by U.S. National Science Foundation grants (including one for the Institute for Artificial Intelligence and Fundamental Interactions), the Singapore Defense Science and Technology Agency, and by funding from Gwangju Institute of Science and Technology and Amazon. Their work will be presented at the Conference on Computer Vision and Pattern Recognition in June. More

    • 50 Shares129 Views

      in Data Management & Statistics

      Exploring the cellular neighborhood

      by

      Cells rely on complex molecular machines composed of protein assemblies to perform essential functions such as energy production, gene expression, and protein synthesis. To better understand how these machines work, scientists capture snapshots of them by isolating proteins from cells and using various methods to determine their structures. However, isolating proteins from cells also removes them from the context of their native environment, including protein interaction partners and cellular location.

      Recently, cryogenic electron tomography (cryo-ET) has emerged as a way to observe proteins in their native environment by imaging frozen cells at different angles to obtain three-dimensional structural information. This approach is exciting because it allows researchers to directly observe how and where proteins associate with each other, revealing the cellular neighborhood of those interactions within the cell.

      With the technology available to image proteins in their native environment, MIT graduate student Barrett Powell wondered if he could take it one step further: What if molecular machines could be observed in action? In a paper published March 8 in Nature Methods, Powell describes the method he developed, called tomoDRGN, for modeling structural differences of proteins in cryo-ET data that arise from protein motions or proteins binding to different interaction partners. These variations are known as structural heterogeneity. 

      Although Powell had joined the lab of MIT associate professor of biology Joey Davis as an experimental scientist, he recognized the potential impact of computational approaches in understanding structural heterogeneity within a cell. Previously, the Davis Lab developed a related methodology named cryoDRGN to understand structural heterogeneity in purified samples. As Powell and Davis saw cryo-ET rising in prominence in the field, Powell took on the challenge of re-imagining this framework to work in cells.

      When solving structures with purified samples, each particle is imaged only once. By contrast, cryo-ET data is collected by imaging each particle more than 40 times from different angles. That meant tomoDRGN needed to be able to merge the information from more than 40 images, which was where the project hit a roadblock: the amount of data led to an information overload.

      To address this, Powell successfully rebuilt the cryoDRGN model to prioritize only the highest-quality data. When imaging the same particle multiple times, radiation damage occurs. The images acquired earlier, therefore, tend to be of higher quality because the particles are less damaged.

      “By excluding some of the lower-quality data, the results were actually better than using all of the data — and the computational performance was substantially faster,” Powell says.

      Just as Powell was beginning work on testing his model, he had a stroke of luck: The authors of a groundbreaking new study that visualized, for the first time, ribosomes inside cells at near-atomic resolution, shared their raw data on the Electric Microscopy Public Image Archive (EMPIAR). This dataset was an exemplary test case for Powell, through which he demonstrated that tomoDRGN could uncover structural heterogeneity within cryo-ET data. 

      According to Powell, one exciting result is what tomoDRGN found surrounding a subset of ribosomes in the EMPIAR dataset. Some of the ribosomal particles were associated with a bacterial cell membrane and engaged in a process called cotranslational translocation. This occurs when a protein is being simultaneously synthesized and transported across a membrane. Researchers can use this result to make new hypotheses about how the ribosome functions with other protein machinery integral to transporting proteins outside of the cell, now guided by a structure of the complex in its native environment. 

      After seeing that tomoDRGN could resolve structural heterogeneity from a structurally diverse dataset, Powell was curious: How small of a population could tomoDRGN identify? For that test, he chose a protein named apoferritin, which is a commonly used benchmark for cryo-ET and is often treated as structurally homogeneous. Ferritin is a protein used for iron storage and is referred to as apoferritin when it lacks iron.

      Surprisingly, in addition to the expected particles, tomoDRGN revealed a minor population of ferritin particles — with iron bound — making up just 2 percent of the dataset, that was not previously reported. This result further demonstrated tomoDRGN’s ability to identify structural states that occur so infrequently that they would be averaged out of a 3D reconstruction. 

      Powell and other members of the Davis Lab are excited to see how tomoDRGN can be applied to further ribosomal studies and to other systems. Davis works on understanding how cells assemble, regulate, and degrade molecular machines, so the next steps include exploring ribosome biogenesis within cells in greater detail using this new tool.

      “What are the possible states that we may be losing during purification?” Davis asks. “Perhaps more excitingly, we can look at how they localize within the cell and what partners and protein complexes they may be interacting with.” More

    • 88 Shares149 Views

      in Data Management & Statistics

      Using generative AI to improve software testing

      by

      Generative AI is getting plenty of attention for its ability to create text and images. But those media represent only a fraction of the data that proliferate in our society today. Data are generated every time a patient goes through a medical system, a storm impacts a flight, or a person interacts with a software application.

      Using generative AI to create realistic synthetic data around those scenarios can help organizations more effectively treat patients, reroute planes, or improve software platforms — especially in scenarios where real-world data are limited or sensitive.

      For the last three years, the MIT spinout DataCebo has offered a generative software system called the Synthetic Data Vault to help organizations create synthetic data to do things like test software applications and train machine learning models.

      The Synthetic Data Vault, or SDV, has been downloaded more than 1 million times, with more than 10,000 data scientists using the open-source library for generating synthetic tabular data. The founders — Principal Research Scientist Kalyan Veeramachaneni and alumna Neha Patki ’15, SM ’16 — believe the company’s success is due to SDV’s ability to revolutionize software testing.

      SDV goes viral

      In 2016, Veeramachaneni’s group in the Data to AI Lab unveiled a suite of open-source generative AI tools to help organizations create synthetic data that matched the statistical properties of real data.

      Companies can use synthetic data instead of sensitive information in programs while still preserving the statistical relationships between datapoints. Companies can also use synthetic data to run new software through simulations to see how it performs before releasing it to the public.

      Veeramachaneni’s group came across the problem because it was working with companies that wanted to share their data for research.

      “MIT helps you see all these different use cases,” Patki explains. “You work with finance companies and health care companies, and all those projects are useful to formulate solutions across industries.”

      In 2020, the researchers founded DataCebo to build more SDV features for larger organizations. Since then, the use cases have been as impressive as they’ve been varied.

      With DataCebo’s new flight simulator, for instance, airlines can plan for rare weather events in a way that would be impossible using only historic data. In another application, SDV users synthesized medical records to predict health outcomes for patients with cystic fibrosis. A team from Norway recently used SDV to create synthetic student data to evaluate whether various admissions policies were meritocratic and free from bias.

      In 2021, the data science platform Kaggle hosted a competition for data scientists that used SDV to create synthetic data sets to avoid using proprietary data. Roughly 30,000 data scientists participated, building solutions and predicting outcomes based on the company’s realistic data.

      And as DataCebo has grown, it’s stayed true to its MIT roots: All of the company’s current employees are MIT alumni.

      Supercharging software testing

      Although their open-source tools are being used for a variety of use cases, the company is focused on growing its traction in software testing.

      “You need data to test these software applications,” Veeramachaneni says. “Traditionally, developers manually write scripts to create synthetic data. With generative models, created using SDV, you can learn from a sample of data collected and then sample a large volume of synthetic data (which has the same properties as real data), or create specific scenarios and edge cases, and use the data to test your application.”

      For example, if a bank wanted to test a program designed to reject transfers from accounts with no money in them, it would have to simulate many accounts simultaneously transacting. Doing that with data created manually would take a lot of time. With DataCebo’s generative models, customers can create any edge case they want to test.

      “It’s common for industries to have data that is sensitive in some capacity,” Patki says. “Often when you’re in a domain with sensitive data you’re dealing with regulations, and even if there aren’t legal regulations, it’s in companies’ best interest to be diligent about who gets access to what at which time. So, synthetic data is always better from a privacy perspective.”

      Scaling synthetic data

      Veeramachaneni believes DataCebo is advancing the field of what it calls synthetic enterprise data, or data generated from user behavior on large companies’ software applications.

      “Enterprise data of this kind is complex, and there is no universal availability of it, unlike language data,” Veeramachaneni says. “When folks use our publicly available software and report back if works on a certain pattern, we learn a lot of these unique patterns, and it allows us to improve our algorithms. From one perspective, we are building a corpus of these complex patterns, which for language and images is readily available. “

      DataCebo also recently released features to improve SDV’s usefulness, including tools to assess the “realism” of the generated data, called the SDMetrics library as well as a way to compare models’ performances called SDGym.

      “It’s about ensuring organizations trust this new data,” Veeramachaneni says. “[Our tools offer] programmable synthetic data, which means we allow enterprises to insert their specific insight and intuition to build more transparent models.”

      As companies in every industry rush to adopt AI and other data science tools, DataCebo is ultimately helping them do so in a way that is more transparent and responsible.

      “In the next few years, synthetic data from generative models will transform all data work,” Veeramachaneni says. “We believe 90 percent of enterprise operations can be done with synthetic data.” More

    • 125 Shares199 Views

      in Data Management & Statistics

      New AI model could streamline operations in a robotic warehouse

      by

      Hundreds of robots zip back and forth across the floor of a colossal robotic warehouse, grabbing items and delivering them to human workers for packing and shipping. Such warehouses are increasingly becoming part of the supply chain in many industries, from e-commerce to automotive production.

      However, getting 800 robots to and from their destinations efficiently while keeping them from crashing into each other is no easy task. It is such a complex problem that even the best path-finding algorithms struggle to keep up with the breakneck pace of e-commerce or manufacturing. 

      In a sense, these robots are like cars trying to navigate a crowded city center. So, a group of MIT researchers who use AI to mitigate traffic congestion applied ideas from that domain to tackle this problem.

      They built a deep-learning model that encodes important information about the warehouse, including the robots, planned paths, tasks, and obstacles, and uses it to predict the best areas of the warehouse to decongest to improve overall efficiency.

      Their technique divides the warehouse robots into groups, so these smaller groups of robots can be decongested faster with traditional algorithms used to coordinate robots. In the end, their method decongests the robots nearly four times faster than a strong random search method.

      In addition to streamlining warehouse operations, this deep learning approach could be used in other complex planning tasks, like computer chip design or pipe routing in large buildings.

      “We devised a new neural network architecture that is actually suitable for real-time operations at the scale and complexity of these warehouses. It can encode hundreds of robots in terms of their trajectories, origins, destinations, and relationships with other robots, and it can do this in an efficient manner that reuses computation across groups of robots,” says Cathy Wu, the Gilbert W. Winslow Career Development Assistant Professor in Civil and Environmental Engineering (CEE), and a member of a member of the Laboratory for Information and Decision Systems (LIDS) and the Institute for Data, Systems, and Society (IDSS).

      Wu, senior author of a paper on this technique, is joined by lead author Zhongxia Yan, a graduate student in electrical engineering and computer science. The work will be presented at the International Conference on Learning Representations.

      Robotic Tetris

      From a bird’s eye view, the floor of a robotic e-commerce warehouse looks a bit like a fast-paced game of “Tetris.”

      When a customer order comes in, a robot travels to an area of the warehouse, grabs the shelf that holds the requested item, and delivers it to a human operator who picks and packs the item. Hundreds of robots do this simultaneously, and if two robots’ paths conflict as they cross the massive warehouse, they might crash.

      Traditional search-based algorithms avoid potential crashes by keeping one robot on its course and replanning a trajectory for the other. But with so many robots and potential collisions, the problem quickly grows exponentially.

      “Because the warehouse is operating online, the robots are replanned about every 100 milliseconds. That means that every second, a robot is replanned 10 times. So, these operations need to be very fast,” Wu says.

      Because time is so critical during replanning, the MIT researchers use machine learning to focus the replanning on the most actionable areas of congestion — where there exists the most potential to reduce the total travel time of robots.

      Wu and Yan built a neural network architecture that considers smaller groups of robots at the same time. For instance, in a warehouse with 800 robots, the network might cut the warehouse floor into smaller groups that contain 40 robots each.

      Then, it predicts which group has the most potential to improve the overall solution if a search-based solver were used to coordinate trajectories of robots in that group.

      An iterative process, the overall algorithm picks the most promising robot group with the neural network, decongests the group with the search-based solver, then picks the next most promising group with the neural network, and so on.

      Considering relationships

      The neural network can reason about groups of robots efficiently because it captures complicated relationships that exist between individual robots. For example, even though one robot may be far away from another initially, their paths could still cross during their trips.

      The technique also streamlines computation by encoding constraints only once, rather than repeating the process for each subproblem. For instance, in a warehouse with 800 robots, decongesting a group of 40 robots requires holding the other 760 robots as constraints. Other approaches require reasoning about all 800 robots once per group in each iteration.

      Instead, the researchers’ approach only requires reasoning about the 800 robots once across all groups in each iteration.

      “The warehouse is one big setting, so a lot of these robot groups will have some shared aspects of the larger problem. We designed our architecture to make use of this common information,” she adds.

      They tested their technique in several simulated environments, including some set up like warehouses, some with random obstacles, and even maze-like settings that emulate building interiors.

      By identifying more effective groups to decongest, their learning-based approach decongests the warehouse up to four times faster than strong, non-learning-based approaches. Even when they factored in the additional computational overhead of running the neural network, their approach still solved the problem 3.5 times faster.

      In the future, the researchers want to derive simple, rule-based insights from their neural model, since the decisions of the neural network can be opaque and difficult to interpret. Simpler, rule-based methods could also be easier to implement and maintain in actual robotic warehouse settings.

      “This approach is based on a novel architecture where convolution and attention mechanisms interact effectively and efficiently. Impressively, this leads to being able to take into account the spatiotemporal component of the constructed paths without the need of problem-specific feature engineering. The results are outstanding: Not only is it possible to improve on state-of-the-art large neighborhood search methods in terms of quality of the solution and speed, but the model generalizes to unseen cases wonderfully,” says Andrea Lodi, the Andrew H. and Ann R. Tisch Professor at Cornell Tech, and who was not involved with this research.

      This work was supported by Amazon and the MIT Amazon Science Hub. More