More stories

  • in

    On the road to cleaner, greener, and faster driving

    No one likes sitting at a red light. But signalized intersections aren’t just a minor nuisance for drivers; vehicles consume fuel and emit greenhouse gases while waiting for the light to change.

    What if motorists could time their trips so they arrive at the intersection when the light is green? While that might be just a lucky break for a human driver, it could be achieved more consistently by an autonomous vehicle that uses artificial intelligence to control its speed.

    In a new study, MIT researchers demonstrate a machine-learning approach that can learn to control a fleet of autonomous vehicles as they approach and travel through a signalized intersection in a way that keeps traffic flowing smoothly.

    Using simulations, they found that their approach reduces fuel consumption and emissions while improving average vehicle speed. The technique gets the best results if all cars on the road are autonomous, but even if only 25 percent use their control algorithm, it still leads to substantial fuel and emissions benefits.

    “This is a really interesting place to intervene. No one’s life is better because they were stuck at an intersection. With a lot of other climate change interventions, there is a quality-of-life difference that is expected, so there is a barrier to entry there. Here, the barrier is much lower,” says senior author Cathy Wu, the Gilbert W. Winslow Career Development Assistant Professor in the Department of Civil and Environmental Engineering and a member of the Institute for Data, Systems, and Society (IDSS) and the Laboratory for Information and Decision Systems (LIDS).

    The lead author of the study is Vindula Jayawardana, a graduate student in LIDS and the Department of Electrical Engineering and Computer Science. The research will be presented at the European Control Conference.

    Intersection intricacies

    While humans may drive past a green light without giving it much thought, intersections can present billions of different scenarios depending on the number of lanes, how the signals operate, the number of vehicles and their speeds, the presence of pedestrians and cyclists, etc.

    Typical approaches for tackling intersection control problems use mathematical models to solve one simple, ideal intersection. That looks good on paper, but likely won’t hold up in the real world, where traffic patterns are often about as messy as they come.

    Wu and Jayawardana shifted gears and approached the problem using a model-free technique known as deep reinforcement learning. Reinforcement learning is a trial-and-error method where the control algorithm learns to make a sequence of decisions. It is rewarded when it finds a good sequence. With deep reinforcement learning, the algorithm leverages assumptions learned by a neural network to find shortcuts to good sequences, even if there are billions of possibilities.

    This is useful for solving a long-horizon problem like this; the control algorithm must issue upwards of 500 acceleration instructions to a vehicle over an extended time period, Wu explains.

    “And we have to get the sequence right before we know that we have done a good job of mitigating emissions and getting to the intersection at a good speed,” she adds.

    But there’s an additional wrinkle. The researchers want the system to learn a strategy that reduces fuel consumption and limits the impact on travel time. These goals can be conflicting.

    “To reduce travel time, we want the car to go fast, but to reduce emissions, we want the car to slow down or not move at all. Those competing rewards can be very confusing to the learning agent,” Wu says.

    While it is challenging to solve this problem in its full generality, the researchers employed a workaround using a technique known as reward shaping. With reward shaping, they give the system some domain knowledge it is unable to learn on its own. In this case, they penalized the system whenever the vehicle came to a complete stop, so it would learn to avoid that action.

    Traffic tests

    Once they developed an effective control algorithm, they evaluated it using a traffic simulation platform with a single intersection. The control algorithm is applied to a fleet of connected autonomous vehicles, which can communicate with upcoming traffic lights to receive signal phase and timing information and observe their immediate surroundings. The control algorithm tells each vehicle how to accelerate and decelerate.

    Their system didn’t create any stop-and-go traffic as vehicles approached the intersection. (Stop-and-go traffic occurs when cars are forced to come to a complete stop due to stopped traffic ahead). In simulations, more cars made it through in a single green phase, which outperformed a model that simulates human drivers. When compared to other optimization methods also designed to avoid stop-and-go traffic, their technique resulted in larger fuel consumption and emissions reductions. If every vehicle on the road is autonomous, their control system can reduce fuel consumption by 18 percent and carbon dioxide emissions by 25 percent, while boosting travel speeds by 20 percent.

    “A single intervention having 20 to 25 percent reduction in fuel or emissions is really incredible. But what I find interesting, and was really hoping to see, is this non-linear scaling. If we only control 25 percent of vehicles, that gives us 50 percent of the benefits in terms of fuel and emissions reduction. That means we don’t have to wait until we get to 100 percent autonomous vehicles to get benefits from this approach,” she says.

    Down the road, the researchers want to study interaction effects between multiple intersections. They also plan to explore how different intersection set-ups (number of lanes, signals, timings, etc.) can influence travel time, emissions, and fuel consumption. In addition, they intend to study how their control system could impact safety when autonomous vehicles and human drivers share the road. For instance, even though autonomous vehicles may drive differently than human drivers, slower roadways and roadways with more consistent speeds could improve safety, Wu says.

    While this work is still in its early stages, Wu sees this approach as one that could be more feasibly implemented in the near-term.

    “The aim in this work is to move the needle in sustainable mobility. We want to dream, as well, but these systems are big monsters of inertia. Identifying points of intervention that are small changes to the system but have significant impact is something that gets me up in the morning,” she says.  

    This work was supported, in part, by the MIT-IBM Watson AI Lab. More

  • in

    Technique protects privacy when making online recommendations

    Algorithms recommend products while we shop online or suggest songs we might like as we listen to music on streaming apps.

    These algorithms work by using personal information like our past purchases and browsing history to generate tailored recommendations. The sensitive nature of such data makes preserving privacy extremely important, but existing methods for solving this problem rely on heavy cryptographic tools requiring enormous amounts of computation and bandwidth.

    MIT researchers may have a better solution. They developed a privacy-preserving protocol that is so efficient it can run on a smartphone over a very slow network. Their technique safeguards personal data while ensuring recommendation results are accurate.

    In addition to user privacy, their protocol minimizes the unauthorized transfer of information from the database, known as leakage, even if a malicious agent tries to trick a database into revealing secret information.

    The new protocol could be especially useful in situations where data leaks could violate user privacy laws, like when a health care provider uses a patient’s medical history to search a database for other patients who had similar symptoms or when a company serves targeted advertisements to users under European privacy regulations.

    “This is a really hard problem. We relied on a whole string of cryptographic and algorithmic tricks to arrive at our protocol,” says Sacha Servan-Schreiber, a graduate student in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and lead author of the paper that presents this new protocol.

    Servan-Schreiber wrote the paper with fellow CSAIL graduate student Simon Langowski and their advisor and senior author Srinivas Devadas, the Edwin Sibley Webster Professor of Electrical Engineering. The research will be presented at the IEEE Symposium on Security and Privacy.

    The data next door

    The technique at the heart of algorithmic recommendation engines is known as a nearest neighbor search, which involves finding the data point in a database that is closest to a query point. Data points that are mapped nearby share similar attributes and are called neighbors.

    These searches involve a server that is linked with an online database which contains concise representations of data point attributes. In the case of a music streaming service, those attributes, known as feature vectors, could be the genre or popularity of different songs.

    To find a song recommendation, the client (user) sends a query to the server that contains a certain feature vector, like a genre of music the user likes or a compressed history of their listening habits. The server then provides the ID of a feature vector in the database that is closest to the client’s query, without revealing the actual vector. In the case of music streaming, that ID would likely be a song title. The client learns the recommended song title without learning the feature vector associated with it.

    “The server has to be able to do this computation without seeing the numbers it is doing the computation on. It can’t actually see the features, but still needs to give you the closest thing in the database,” says Langowski.

    To achieve this, the researchers created a protocol that relies on two separate servers that access the same database. Using two servers makes the process more efficient and enables the use of a cryptographic technique known as private information retrieval. This technique allows a client to query a database without revealing what it is searching for, Servan-Schreiber explains.

    Overcoming security challenges

    But while private information retrieval is secure on the client side, it doesn’t provide database privacy on its own. The database offers a set of candidate vectors — possible nearest neighbors — for the client, which are typically winnowed down later by the client using brute force. However, doing so can reveal a lot about the database to the client. The additional privacy challenge is to prevent the client from learning those extra vectors. 

    The researchers employed a tuning technique that eliminates many of the extra vectors in the first place, and then used a different trick, which they call oblivious masking, to hide any additional data points except for the actual nearest neighbor. This efficiently preserves database privacy, so the client won’t learn anything about the feature vectors in the database.  

    Once they designed this protocol, they tested it with a nonprivate implementation on four real-world datasets to determine how to tune the algorithm to maximize accuracy. Then, they used their protocol to conduct private nearest neighbor search queries on those datasets.

    Their technique requires a few seconds of server processing time per query and less than 10 megabytes of communication between the client and servers, even with databases that contained more than 10 million items. By contrast, other secure methods can require gigabytes of communication or hours of computation time. With each query, their method achieved greater than 95 percent accuracy (meaning that nearly every time it found the actual approximate nearest neighbor to the query point). 

    The techniques they used to enable database privacy will thwart a malicious client even if it sends false queries to try and trick the server into leaking information.

    “A malicious client won’t learn much more information than an honest client following protocol. And it protects against malicious servers, too. If one deviates from protocol, you might not get the right result, but they will never learn what the client’s query was,” Langowski says.

    In the future, the researchers plan to adjust the protocol so it can preserve privacy using only one server. This could enable it to be applied in more real-world situations, since it would not require the use of two noncolluding entities (which don’t share information with each other) to manage the database.  

    “Nearest neighbor search undergirds many critical machine-learning driven applications, from providing users with content recommendations to classifying medical conditions. However, it typically requires sharing a lot of data with a central system to aggregate and enable the search,” says Bayan Bruss, head of applied machine-learning research at Capital One, who was not involved with this work. “This research provides a key step towards ensuring that the user receives the benefits from nearest neighbor search while having confidence that the central system will not use their data for other purposes.” More

  • in

    Artificial intelligence system learns concepts shared across video, audio, and text

    Humans observe the world through a combination of different modalities, like vision, hearing, and our understanding of language. Machines, on the other hand, interpret the world through data that algorithms can process.

    So, when a machine “sees” a photo, it must encode that photo into data it can use to perform a task like image classification. This process becomes more complicated when inputs come in multiple formats, like videos, audio clips, and images.

    “The main challenge here is, how can a machine align those different modalities? As humans, this is easy for us. We see a car and then hear the sound of a car driving by, and we know these are the same thing. But for machine learning, it is not that straightforward,” says Alexander Liu, a graduate student in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and first author of a paper tackling this problem. 

    Liu and his collaborators developed an artificial intelligence technique that learns to represent data in a way that captures concepts which are shared between visual and audio modalities. For instance, their method can learn that the action of a baby crying in a video is related to the spoken word “crying” in an audio clip.

    Using this knowledge, their machine-learning model can identify where a certain action is taking place in a video and label it.

    It performs better than other machine-learning methods at cross-modal retrieval tasks, which involve finding a piece of data, like a video, that matches a user’s query given in another form, like spoken language. Their model also makes it easier for users to see why the machine thinks the video it retrieved matches their query.

    This technique could someday be utilized to help robots learn about concepts in the world through perception, more like the way humans do.

    Joining Liu on the paper are CSAIL postdoc SouYoung Jin; grad students Cheng-I Jeff Lai and Andrew Rouditchenko; Aude Oliva, senior research scientist in CSAIL and MIT director of the MIT-IBM Watson AI Lab; and senior author James Glass, senior research scientist and head of the Spoken Language Systems Group in CSAIL. The research will be presented at the Annual Meeting of the Association for Computational Linguistics.

    Learning representations

    The researchers focus their work on representation learning, which is a form of machine learning that seeks to transform input data to make it easier to perform a task like classification or prediction.

    The representation learning model takes raw data, such as videos and their corresponding text captions, and encodes them by extracting features, or observations about objects and actions in the video. Then it maps those data points in a grid, known as an embedding space. The model clusters similar data together as single points in the grid. Each of these data points, or vectors, is represented by an individual word.

    For instance, a video clip of a person juggling might be mapped to a vector labeled “juggling.”

    The researchers constrain the model so it can only use 1,000 words to label vectors. The model can decide which actions or concepts it wants to encode into a single vector, but it can only use 1,000 vectors. The model chooses the words it thinks best represent the data.

    Rather than encoding data from different modalities onto separate grids, their method employs a shared embedding space where two modalities can be encoded together. This enables the model to learn the relationship between representations from two modalities, like video that shows a person juggling and an audio recording of someone saying “juggling.”

    To help the system process data from multiple modalities, they designed an algorithm that guides the machine to encode similar concepts into the same vector.

    “If there is a video about pigs, the model might assign the word ‘pig’ to one of the 1,000 vectors. Then if the model hears someone saying the word ‘pig’ in an audio clip, it should still use the same vector to encode that,” Liu explains.

    A better retriever

    They tested the model on cross-modal retrieval tasks using three datasets: a video-text dataset with video clips and text captions, a video-audio dataset with video clips and spoken audio captions, and an image-audio dataset with images and spoken audio captions.

    For example, in the video-audio dataset, the model chose 1,000 words to represent the actions in the videos. Then, when the researchers fed it audio queries, the model tried to find the clip that best matched those spoken words.

    “Just like a Google search, you type in some text and the machine tries to tell you the most relevant things you are searching for. Only we do this in the vector space,” Liu says.

    Not only was their technique more likely to find better matches than the models they compared it to, it is also easier to understand.

    Because the model could only use 1,000 total words to label vectors, a user can more see easily which words the machine used to conclude that the video and spoken words are similar. This could make the model easier to apply in real-world situations where it is vital that users understand how it makes decisions, Liu says.

    The model still has some limitations they hope to address in future work. For one, their research focused on data from two modalities at a time, but in the real world humans encounter many data modalities simultaneously, Liu says.

    “And we know 1,000 words works on this kind of dataset, but we don’t know if it can be generalized to a real-world problem,” he adds.

    Plus, the images and videos in their datasets contained simple objects or straightforward actions; real-world data are much messier. They also want to determine how well their method scales up when there is a wider diversity of inputs.

    This research was supported, in part, by the MIT-IBM Watson AI Lab and its member companies, Nexplore and Woodside, and by the MIT Lincoln Laboratory. More

  • in

    Engineers use artificial intelligence to capture the complexity of breaking waves

    Waves break once they swell to a critical height, before cresting and crashing into a spray of droplets and bubbles. These waves can be as large as a surfer’s point break and as small as a gentle ripple rolling to shore. For decades, the dynamics of how and when a wave breaks have been too complex to predict.

    Now, MIT engineers have found a new way to model how waves break. The team used machine learning along with data from wave-tank experiments to tweak equations that have traditionally been used to predict wave behavior. Engineers typically rely on such equations to help them design resilient offshore platforms and structures. But until now, the equations have not been able to capture the complexity of breaking waves.

    The updated model made more accurate predictions of how and when waves break, the researchers found. For instance, the model estimated a wave’s steepness just before breaking, and its energy and frequency after breaking, more accurately than the conventional wave equations.

    Their results, published today in the journal Nature Communications, will help scientists understand how a breaking wave affects the water around it. Knowing precisely how these waves interact can help hone the design of offshore structures. It can also improve predictions for how the ocean interacts with the atmosphere. Having better estimates of how waves break can help scientists predict, for instance, how much carbon dioxide and other atmospheric gases the ocean can absorb.

    “Wave breaking is what puts air into the ocean,” says study author Themis Sapsis, an associate professor of mechanical and ocean engineering and an affiliate of the Institute for Data, Systems, and Society at MIT. “It may sound like a detail, but if you multiply its effect over the area of the entire ocean, wave breaking starts becoming fundamentally important to climate prediction.”

    The study’s co-authors include lead author and MIT postdoc Debbie Eeltink, Hubert Branger and Christopher Luneau of Aix-Marseille University, Amin Chabchoub of Kyoto University, Jerome Kasparian of the University of Geneva, and T.S. van den Bremer of Delft University of Technology.

    Learning tank

    To predict the dynamics of a breaking wave, scientists typically take one of two approaches: They either attempt to precisely simulate the wave at the scale of individual molecules of water and air, or they run experiments to try and characterize waves with actual measurements. The first approach is computationally expensive and difficult to simulate even over a small area; the second requires a huge amount of time to run enough experiments to yield statistically significant results.

    The MIT team instead borrowed pieces from both approaches to develop a more efficient and accurate model using machine learning. The researchers started with a set of equations that is considered the standard description of wave behavior. They aimed to improve the model by “training” the model on data of breaking waves from actual experiments.

    “We had a simple model that doesn’t capture wave breaking, and then we had the truth, meaning experiments that involve wave breaking,” Eeltink explains. “Then we wanted to use machine learning to learn the difference between the two.”

    The researchers obtained wave breaking data by running experiments in a 40-meter-long tank. The tank was fitted at one end with a paddle which the team used to initiate each wave. The team set the paddle to produce a breaking wave in the middle of the tank. Gauges along the length of the tank measured the water’s height as waves propagated down the tank.

    “It takes a lot of time to run these experiments,” Eeltink says. “Between each experiment you have to wait for the water to completely calm down before you launch the next experiment, otherwise they influence each other.”

    Safe harbor

    In all, the team ran about 250 experiments, the data from which they used to train a type of machine-learning algorithm known as a neural network. Specifically, the algorithm is trained to compare the real waves in experiments with the predicted waves in the simple model, and based on any differences between the two, the algorithm tunes the model to fit reality.

    After training the algorithm on their experimental data, the team introduced the model to entirely new data — in this case, measurements from two independent experiments, each run at separate wave tanks with different dimensions. In these tests, they found the updated model made more accurate predictions than the simple, untrained model, for instance making better estimates of a breaking wave’s steepness.

    The new model also captured an essential property of breaking waves known as the “downshift,” in which the frequency of a wave is shifted to a lower value. The speed of a wave depends on its frequency. For ocean waves, lower frequencies move faster than higher frequencies. Therefore, after the downshift, the wave will move faster. The new model predicts the change in frequency, before and after each breaking wave, which could be especially relevant in preparing for coastal storms.

    “When you want to forecast when high waves of a swell would reach a harbor, and you want to leave the harbor before those waves arrive, then if you get the wave frequency wrong, then the speed at which the waves are approaching is wrong,” Eeltink says.

    The team’s updated wave model is in the form of an open-source code that others could potentially use, for instance in climate simulations of the ocean’s potential to absorb carbon dioxide and other atmospheric gases. The code can also be worked into simulated tests of offshore platforms and coastal structures.

    “The number one purpose of this model is to predict what a wave will do,” Sapsis says. “If you don’t model wave breaking right, it would have tremendous implications for how structures behave. With this, you could simulate waves to help design structures better, more efficiently, and without huge safety factors.”

    This research is supported, in part, by the Swiss National Science Foundation, and by the U.S. Office of Naval Research. More

  • in

    Estimating the informativeness of data

    Not all data are created equal. But how much information is any piece of data likely to contain? This question is central to medical testing, designing scientific experiments, and even to everyday human learning and thinking. MIT researchers have developed a new way to solve this problem, opening up new applications in medicine, scientific discovery, cognitive science, and artificial intelligence.

    In theory, the 1948 paper, “A Mathematical Theory of Communication,” by the late MIT Professor Emeritus Claude Shannon answered this question definitively. One of Shannon’s breakthrough results is the idea of entropy, which lets us quantify the amount of information inherent in any random object, including random variables that model observed data. Shannon’s results created the foundations of information theory and modern telecommunications. The concept of entropy has also proven central to computer science and machine learning.

    The challenge of estimating entropy

    Unfortunately, the use of Shannon’s formula can quickly become computationally intractable. It requires precisely calculating the probability of the data, which in turn requires calculating every possible way the data could have arisen under a probabilistic model. If the data-generating process is very simple — for example, a single toss of a coin or roll of a loaded die — then calculating entropies is straightforward. But consider the problem of medical testing, where a positive test result is the result of hundreds of interacting variables, all unknown. With just 10 unknowns, there are already 1,000 possible explanations for the data. With a few hundred, there are more possible explanations than atoms in the known universe, which makes calculating the entropy exactly an unmanageable problem.

    MIT researchers have developed a new method to estimate good approximations to many information quantities such as Shannon entropy by using probabilistic inference. The work appears in a paper presented at AISTATS 2022 by authors Feras Saad ’16, MEng ’16, a PhD candidate in electrical engineering and computer science; Marco-Cusumano Towner PhD ’21; and Vikash Mansinghka ’05, MEng ’09, PhD ’09, a principal research scientist in the Department of Brain and Cognitive Sciences. The key insight is, rather than enumerate all explanations, to instead use probabilistic inference algorithms to first infer which explanations are probable and then use these probable explanations to construct high-quality entropy estimates. The paper shows that this inference-based approach can be much faster and more accurate than previous approaches.

    Estimating entropy and information in a probabilistic model is fundamentally hard because it often requires solving a high-dimensional integration problem. Many previous works have developed estimators of these quantities for certain special cases, but the new estimators of entropy via inference (EEVI) offer the first approach that can deliver sharp upper and lower bounds on a broad set of information-theoretic quantities. An upper and lower bound means that although we don’t know the true entropy, we can get a number that is smaller than it and a number that is higher than it.

    “The upper and lower bounds on entropy delivered by our method are particularly useful for three reasons,” says Saad. “First, the difference between the upper and lower bounds gives a quantitative sense of how confident we should be about the estimates. Second, by using more computational effort we can drive the difference between the two bounds to zero, which ‘squeezes’ the true value with a high degree of accuracy. Third, we can compose these bounds to form estimates of many other quantities that tell us how informative different variables in a model are of one another.”

    Solving fundamental problems with data-driven expert systems

    Saad says he is most excited about the possibility that this method gives for querying probabilistic models in areas like machine-assisted medical diagnoses. He says one goal of the EEVI method is to be able to solve new queries using rich generative models for things like liver disease and diabetes that have already been developed by experts in the medical domain. For example, suppose we have a patient with a set of observed attributes (height, weight, age, etc.) and observed symptoms (nausea, blood pressure, etc.). Given these attributes and symptoms, EEVI can be used to help determine which medical tests for symptoms the physician should conduct to maximize information about the absence or presence of a given liver disease (like cirrhosis or primary biliary cholangitis).

    For insulin diagnosis, the authors showed how to use the method for computing optimal times to take blood glucose measurements that maximize information about a patient’s insulin sensitivity, given an expert-built probabilistic model of insulin metabolism and the patient’s personalized meal and medication schedule. As routine medical tracking like glucose monitoring moves away from doctor’s offices and toward wearable devices, there are even more opportunities to improve data acquisition, if the value of the data can be estimated accurately in advance.

    Vikash Mansinghka, senior author on the paper, adds, “We’ve shown that probabilistic inference algorithms can be used to estimate rigorous bounds on information measures that AI engineers often think of as intractable to calculate. This opens up many new applications. It also shows that inference may be more computationally fundamental than we thought. It also helps to explain how human minds might be able to estimate the value of information so pervasively, as a central building block of everyday cognition, and help us engineer AI expert systems that have these capabilities.”

    The paper, “Estimators of Entropy and Information via Inference in Probabilistic Models,” was presented at AISTATS 2022. More

  • in

    A new state of the art for unsupervised vision

    Labeling data can be a chore. It’s the main source of sustenance for computer-vision models; without it, they’d have a lot of difficulty identifying objects, people, and other important image characteristics. Yet producing just an hour of tagged and labeled data can take a whopping 800 hours of human time. Our high-fidelity understanding of the world develops as machines can better perceive and interact with our surroundings. But they need more help.

    Scientists from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Microsoft, and Cornell University have attempted to solve this problem plaguing vision models by creating “STEGO,” an algorithm that can jointly discover and segment objects without any human labels at all, down to the pixel.

    STEGO learns something called “semantic segmentation” — fancy speak for the process of assigning a label to every pixel in an image. Semantic segmentation is an important skill for today’s computer-vision systems because images can be cluttered with objects. Even more challenging is that these objects don’t always fit into literal boxes; algorithms tend to work better for discrete “things” like people and cars as opposed to “stuff” like vegetation, sky, and mashed potatoes. A previous system might simply perceive a nuanced scene of a dog playing in the park as just a dog, but by assigning every pixel of the image a label, STEGO can break the image into its main ingredients: a dog, sky, grass, and its owner.

    Play video

    A new state of the art for unsupervised computer vision

    Assigning every single pixel of the world a label is ambitious — especially without any kind of feedback from humans. The majority of algorithms today get their knowledge from mounds of labeled data, which can take painstaking human-hours to source. Just imagine the excitement of labeling every pixel of 100,000 images! To discover these objects without a human’s helpful guidance, STEGO looks for similar objects that appear throughout a dataset. It then associates these similar objects together to construct a consistent view of the world across all of the images it learns from.

    Seeing the world

    Machines that can “see” are crucial for a wide array of new and emerging technologies like self-driving cars and predictive modeling for medical diagnostics. Since STEGO can learn without labels, it can detect objects in many different domains, even those that humans don’t yet understand fully. 

    “If you’re looking at oncological scans, the surface of planets, or high-resolution biological images, it’s hard to know what objects to look for without expert knowledge. In emerging domains, sometimes even human experts don’t know what the right objects should be,” says Mark Hamilton, a PhD student in electrical engineering and computer science at MIT, research affiliate of MIT CSAIL, software engineer at Microsoft, and lead author on a new paper about STEGO. “In these types of situations where you want to design a method to operate at the boundaries of science, you can’t rely on humans to figure it out before machines do.”

    STEGO was tested on a slew of visual domains spanning general images, driving images, and high-altitude aerial photographs. In each domain, STEGO was able to identify and segment relevant objects that were closely aligned with human judgments. STEGO’s most diverse benchmark was the COCO-Stuff dataset, which is made up of diverse images from all over the world, from indoor scenes to people playing sports to trees and cows. In most cases, the previous state-of-the-art system could capture a low-resolution gist of a scene, but struggled on fine-grained details: A human was a blob, a motorcycle was captured as a person, and it couldn’t recognize any geese. On the same scenes, STEGO doubled the performance of previous systems and discovered concepts like animals, buildings, people, furniture, and many others.

    STEGO not only doubled the performance of prior systems on the COCO-Stuff benchmark, but made similar leaps forward in other visual domains. When applied to driverless car datasets, STEGO successfully segmented out roads, people, and street signs with much higher resolution and granularity than previous systems. On images from space, the system broke down every single square foot of the surface of the Earth into roads, vegetation, and buildings. 

    Connecting the pixels

    STEGO — which stands for “Self-supervised Transformer with Energy-based Graph Optimization” — builds on top of the DINO algorithm, which learned about the world through 14 million images from the ImageNet database. STEGO refines the DINO backbone through a learning process that mimics our own way of stitching together pieces of the world to make meaning. 

    For example, you might consider two images of dogs walking in the park. Even though they’re different dogs, with different owners, in different parks, STEGO can tell (without humans) how each scene’s objects relate to each other. The authors even probe STEGO’s mind to see how each little, brown, furry thing in the images are similar, and likewise with other shared objects like grass and people. By connecting objects across images, STEGO builds a consistent view of the word.

    “The idea is that these types of algorithms can find consistent groupings in a largely automated fashion so we don’t have to do that ourselves,” says Hamilton. “It might have taken years to understand complex visual datasets like biological imagery, but if we can avoid spending 1,000 hours combing through data and labeling it, we can find and discover new information that we might have missed. We hope this will help us understand the visual word in a more empirically grounded way.”

    Looking ahead

    Despite its improvements, STEGO still faces certain challenges. One is that labels can be arbitrary. For example, the labels of the COCO-Stuff dataset distinguish between “food-things” like bananas and chicken wings, and “food-stuff” like grits and pasta. STEGO doesn’t see much of a distinction there. In other cases, STEGO was confused by odd images — like one of a banana sitting on a phone receiver — where the receiver was labeled “foodstuff,” instead of “raw material.” 

    For upcoming work, they’re planning to explore giving STEGO a bit more flexibility than just labeling pixels into a fixed number of classes as things in the real world can sometimes be multiple things at the same time (like “food”, “plant” and “fruit”). The authors hope this will give the algorithm room for uncertainty, trade-offs, and more abstract thinking.

    “In making a general tool for understanding potentially complicated datasets, we hope that this type of an algorithm can automate the scientific process of object discovery from images. There’s a lot of different domains where human labeling would be prohibitively expensive, or humans simply don’t even know the specific structure, like in certain biological and astrophysical domains. We hope that future work enables application to a very broad scope of datasets. Since you don’t need any human labels, we can now start to apply ML tools more broadly,” says Hamilton.

    “STEGO is simple, elegant, and very effective. I consider unsupervised segmentation to be a benchmark for progress in image understanding, and a very difficult problem. The research community has made terrific progress in unsupervised image understanding with the adoption of transformer architectures,” says Andrea Vedaldi, professor of computer vision and machine learning and a co-lead of the Visual Geometry Group at the engineering science department of the University of Oxford. “This research provides perhaps the most direct and effective demonstration of this progress on unsupervised segmentation.” 

    Hamilton wrote the paper alongside MIT CSAIL PhD student Zhoutong Zhang, Assistant Professor Bharath Hariharan of Cornell University, Associate Professor Noah Snavely of Cornell Tech, and MIT professor William T. Freeman. They will present the paper at the 2022 International Conference on Learning Representations (ICLR).  More

  • in

    MIT Schwarzman College of Computing unveils Break Through Tech AI

    Aimed at driving diversity and inclusion in artificial intelligence, the MIT Stephen A. Schwarzman College of Computing is launching Break Through Tech AI, a new program to bridge the talent gap for women and underrepresented genders in AI positions in industry.

    Break Through Tech AI will provide skills-based training, industry-relevant portfolios, and mentoring to qualified undergraduate students in the Greater Boston area in order to position them more competitively for careers in data science, machine learning, and artificial intelligence. The free, 18-month program will also provide each student with a stipend for participation to lower the barrier for those typically unable to engage in an unpaid, extra-curricular educational opportunity.

    “Helping position students from diverse backgrounds to succeed in fields such as data science, machine learning, and artificial intelligence is critical for our society’s future,” says Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing and Henry Ellis Warren Professor of Electrical Engineering and Computer Science. “We look forward to working with students from across the Greater Boston area to provide them with skills and mentorship to help them find careers in this competitive and growing industry.”

    The college is collaborating with Break Through Tech — a national initiative launched by Cornell Tech in 2016 to increase the number of women and underrepresented groups graduating with degrees in computing — to host and administer the program locally. In addition to Boston, the inaugural artificial intelligence and machine learning program will be offered in two other metropolitan areas — one based in New York hosted by Cornell Tech and another in Los Angeles hosted by the University of California at Los Angeles Samueli School of Engineering.

    “Break Through Tech’s success at diversifying who is pursuing computer science degrees and careers has transformed lives and the industry,” says Judith Spitz, executive director of Break Through Tech. “With our new collaborators, we can apply our impactful model to drive inclusion and diversity in artificial intelligence.”

    The new program will kick off this summer at MIT with an eight-week, skills-based online course and in-person lab experience that teaches industry-relevant tools to build real-world AI solutions. Students will learn how to analyze datasets and use several common machine learning libraries to build, train, and implement their own ML models in a business context.

    Following the summer course, students will be matched with machine-learning challenge projects for which they will convene monthly at MIT and work in teams to build solutions and collaborate with an industry advisor or mentor throughout the academic year, resulting in a portfolio of resume-quality work. The participants will also be paired with young professionals in the field to help build their network, prepare their portfolio, practice for interviews, and cultivate workplace skills.

    “Leveraging the college’s strong partnership with industry, Break Through AI will offer unique opportunities to students that will enhance their portfolio in machine learning and AI,” says Asu Ozdaglar, deputy dean of academics of the MIT Schwarzman College of Computing and head of the Department of Electrical Engineering and Computer Science. Ozdaglar, who will be the MIT faculty director of Break Through Tech AI, adds: “The college is committed to making computing inclusive and accessible for all. We’re thrilled to host this program at MIT for the Greater Boston area and to do what we can to help increase diversity in computing fields.”

    Break Through Tech AI is part of the MIT Schwarzman College of Computing’s focus to advance diversity, equity, and inclusion in computing. The college aims to improve and create programs and activities that broaden participation in computing classes and degree programs, increase the diversity of top faculty candidates in computing fields, and ensure that faculty search and graduate admissions processes have diverse slates of candidates and interviews.

    “By engaging in activities like Break Through Tech AI that work to improve the climate for underrepresented groups, we’re taking an important step toward creating more welcoming environments where all members can innovate and thrive,” says Alana Anderson, assistant dean for diversity, equity and inclusion for the Schwarzman College of Computing. More

  • in

    Does this artificial intelligence think like a human?

    In machine learning, understanding why a model makes certain decisions is often just as important as whether those decisions are correct. For instance, a machine-learning model might correctly predict that a skin lesion is cancerous, but it could have done so using an unrelated blip on a clinical photo.

    While tools exist to help experts make sense of a model’s reasoning, often these methods only provide insights on one decision at a time, and each must be manually evaluated. Models are commonly trained using millions of data inputs, making it almost impossible for a human to evaluate enough decisions to identify patterns.

    Now, researchers at MIT and IBM Research have created a method that enables a user to aggregate, sort, and rank these individual explanations to rapidly analyze a machine-learning model’s behavior. Their technique, called Shared Interest, incorporates quantifiable metrics that compare how well a model’s reasoning matches that of a human.

    Shared Interest could help a user easily uncover concerning trends in a model’s decision-making — for example, perhaps the model often becomes confused by distracting, irrelevant features, like background objects in photos. Aggregating these insights could help the user quickly and quantitatively determine whether a model is trustworthy and ready to be deployed in a real-world situation.

    “In developing Shared Interest, our goal is to be able to scale up this analysis process so that you could understand on a more global level what your model’s behavior is,” says lead author Angie Boggust, a graduate student in the Visualization Group of the Computer Science and Artificial Intelligence Laboratory (CSAIL).

    Boggust wrote the paper with her advisor, Arvind Satyanarayan, an assistant professor of computer science who leads the Visualization Group, as well as Benjamin Hoover and senior author Hendrik Strobelt, both of IBM Research. The paper will be presented at the Conference on Human Factors in Computing Systems.

    Boggust began working on this project during a summer internship at IBM, under the mentorship of Strobelt. After returning to MIT, Boggust and Satyanarayan expanded on the project and continued the collaboration with Strobelt and Hoover, who helped deploy the case studies that show how the technique could be used in practice.

    Human-AI alignment

    Shared Interest leverages popular techniques that show how a machine-learning model made a specific decision, known as saliency methods. If the model is classifying images, saliency methods highlight areas of an image that are important to the model when it made its decision. These areas are visualized as a type of heatmap, called a saliency map, that is often overlaid on the original image. If the model classified the image as a dog, and the dog’s head is highlighted, that means those pixels were important to the model when it decided the image contains a dog.

    Shared Interest works by comparing saliency methods to ground-truth data. In an image dataset, ground-truth data are typically human-generated annotations that surround the relevant parts of each image. In the previous example, the box would surround the entire dog in the photo. When evaluating an image classification model, Shared Interest compares the model-generated saliency data and the human-generated ground-truth data for the same image to see how well they align.

    The technique uses several metrics to quantify that alignment (or misalignment) and then sorts a particular decision into one of eight categories. The categories run the gamut from perfectly human-aligned (the model makes a correct prediction and the highlighted area in the saliency map is identical to the human-generated box) to completely distracted (the model makes an incorrect prediction and does not use any image features found in the human-generated box).

    “On one end of the spectrum, your model made the decision for the exact same reason a human did, and on the other end of the spectrum, your model and the human are making this decision for totally different reasons. By quantifying that for all the images in your dataset, you can use that quantification to sort through them,” Boggust explains.

    The technique works similarly with text-based data, where key words are highlighted instead of image regions.

    Rapid analysis

    The researchers used three case studies to show how Shared Interest could be useful to both nonexperts and machine-learning researchers.

    In the first case study, they used Shared Interest to help a dermatologist determine if he should trust a machine-learning model designed to help diagnose cancer from photos of skin lesions. Shared Interest enabled the dermatologist to quickly see examples of the model’s correct and incorrect predictions. Ultimately, the dermatologist decided he could not trust the model because it made too many predictions based on image artifacts, rather than actual lesions.

    “The value here is that using Shared Interest, we are able to see these patterns emerge in our model’s behavior. In about half an hour, the dermatologist was able to make a confident decision of whether or not to trust the model and whether or not to deploy it,” Boggust says.

    In the second case study, they worked with a machine-learning researcher to show how Shared Interest can evaluate a particular saliency method by revealing previously unknown pitfalls in the model. Their technique enabled the researcher to analyze thousands of correct and incorrect decisions in a fraction of the time required by typical manual methods.

    In the third case study, they used Shared Interest to dive deeper into a specific image classification example. By manipulating the ground-truth area of the image, they were able to conduct a what-if analysis to see which image features were most important for particular predictions.   

    The researchers were impressed by how well Shared Interest performed in these case studies, but Boggust cautions that the technique is only as good as the saliency methods it is based upon. If those techniques contain bias or are inaccurate, then Shared Interest will inherit those limitations.

    In the future, the researchers want to apply Shared Interest to different types of data, particularly tabular data which is used in medical records. They also want to use Shared Interest to help improve current saliency techniques. Boggust hopes this research inspires more work that seeks to quantify machine-learning model behavior in ways that make sense to humans.

    This work is funded, in part, by the MIT-IBM Watson AI Lab, the United States Air Force Research Laboratory, and the United States Air Force Artificial Intelligence Accelerator. More