Tag Archives: respond
A few years back, DeepMind’s Demis Hassabis famously prophesized that AI and neuroscience will positively feed into each other in a “virtuous circle.” If realized, this would fundamentally expand our insight into intelligence, both machine and human.
We’ve already seen some proofs of concept, at least in the brain-to-AI direction. For example, memory replay, a biological mechanism that fortifies our memories during sleep, also boosted AI learning when abstractly appropriated into deep learning models. Reinforcement learning, loosely based on our motivation circuits, is now behind some of AI’s most powerful tools.
Hassabis is about to be proven right again.
Last week, two studies independently tapped into the power of ANNs to solve a 70-year-old neuroscience mystery: how does our visual system perceive reality?
The first, published in Cell, used generative networks to evolve DeepDream-like images that hyper-activate complex visual neurons in monkeys. These machine artworks are pure nightmare fuel to the human eye; but together, they revealed a fundamental “visual hieroglyph” that may form a basic rule for how we piece together visual stimuli to process sight into perception.
In the second study, a team used a deep ANN model—one thought to mimic biological vision—to synthesize new patterns tailored to control certain networks of visual neurons in the monkey brain. When directly shown to monkeys, the team found that the machine-generated artworks could reliably activate predicted populations of neurons. Future improved ANN models could allow even better control, giving neuroscientists a powerful noninvasive tool to study the brain. The work was published in Science.
The individual results, though fascinating, aren’t necessarily the point. Rather, they illustrate how scientists are now striving to complete the virtuous circle: tapping AI to probe natural intelligence. Vision is only the beginning—the tools can potentially be expanded into other sensory domains. And the more we understand about natural brains, the better we can engineer artificial ones.
It’s a “great example of leveraging artificial intelligence to study organic intelligence,” commented Dr. Roman Sandler at Kernel.co on Twitter.
ANNs and biological vision have quite the history.
In the late 1950s, the legendary neuroscientist duo David Hubel and Torsten Wiesel became some of the first to use mathematical equations to understand how neurons in the brain work together.
In a series of experiments—many using cats—the team carefully dissected the structure and function of the visual cortex. Using myriads of images, they revealed that vision is processed in a hierarchy: neurons in “earlier” brain regions, those closer to the eyes, tend to activate when they “see” simple patterns such as lines. As we move deeper into the brain, from the early V1 to a nub located slightly behind our ears, the IT cortex, neurons increasingly respond to more complex or abstract patterns, including faces, animals, and objects. The discovery led some scientists to call certain IT neurons “Jennifer Aniston cells,” which fire in response to pictures of the actress regardless of lighting, angle, or haircut. That is, IT neurons somehow extract visual information into the “gist” of things.
That’s not trivial. The complex neural connections that lead to increasing abstraction of what we see into what we think we see—what we perceive—is a central question in machine vision: how can we teach machines to transform numbers encoding stimuli into dots, lines, and angles that eventually form “perceptions” and “gists”? The answer could transform self-driving cars, facial recognition, and other computer vision applications as they learn to better generalize.
Hubel and Wiesel’s Nobel-prize-winning studies heavily influenced the birth of ANNs and deep learning. Much of earlier ANN “feed-forward” model structures are based on our visual system; even today, the idea of increasing layers of abstraction—for perception or reasoning—guide computer scientists to build AI that can better generalize. The early romance between vision and deep learning is perhaps the bond that kicked off our current AI revolution.
It only seems fair that AI would feed back into vision neuroscience.
Hieroglyphs and Controllers
In the Cell study, a team led by Dr. Margaret Livingstone at Harvard Medical School tapped into generative networks to unravel IT neurons’ complex visual alphabet.
Scientists have long known that neurons in earlier visual regions (V1) tend to fire in response to “grating patches” oriented in certain ways. Using a limited set of these patches like letters, V1 neurons can “express a visual sentence” and represent any image, said Dr. Arash Afraz at the National Institute of Health, who was not involved in the study.
But how IT neurons operate remained a mystery. Here, the team used a combination of genetic algorithms and deep generative networks to “evolve” computer art for every studied neuron. In seven monkeys, the team implanted electrodes into various parts of the visual IT region so that they could monitor the activity of a single neuron.
The team showed each monkey an initial set of 40 images. They then picked the top 10 images that stimulated the highest neural activity, and married them to 30 new images to “evolve” the next generation of images. After 250 generations, the technique, XDREAM, generated a slew of images that mashed up contorted face-like shapes with lines, gratings, and abstract shapes.
This image shows the evolution of an optimum image for stimulating a visual neuron in a monkey. Image Credit: Ponce, Xiao, and Schade et al. – Cell.
“The evolved images look quite counter-intuitive,” explained Afraz. Some clearly show detailed structures that resemble natural images, while others show complex structures that can’t be characterized by our puny human brains.
This figure shows natural images (right) and images evolved by neurons in the inferotemporal cortex of a monkey (left). Image Credit: Ponce, Xiao, and Schade et al. – Cell.
“What started to emerge during each experiment were pictures that were reminiscent of shapes in the world but were not actual objects in the world,” said study author Carlos Ponce. “We were seeing something that was more like the language cells use with each other.”
This image was evolved by a neuron in the inferotemporal cortex of a monkey using AI. Image Credit: Ponce, Xiao, and Schade et al. – Cell.
Although IT neurons don’t seem to use a simple letter alphabet, it does rely on a vast array of characters like hieroglyphs or Chinese characters, “each loaded with more information,” said Afraz.
The adaptive nature of XDREAM turns it into a powerful tool to probe the inner workings of our brains—particularly for revealing discrepancies between biology and models.
The Science study, led by Dr. James DiCarlo at MIT, takes a similar approach. Using ANNs to generate new patterns and images, the team was able to selectively predict and independently control neuron populations in a high-level visual region called V4.
“So far, what has been done with these models is predicting what the neural responses would be to other stimuli that they have not seen before,” said study author Dr. Pouya Bashivan. “The main difference here is that we are going one step further and using the models to drive the neurons into desired states.”
It suggests that our current ANN models for visual computation “implicitly capture a great deal of visual knowledge” which we can’t really describe, but which the brain uses to turn vision information into perception, the authors said. By testing AI-generated images on biological vision, however, the team concluded that today’s ANNs have a degree of understanding and generalization. The results could potentially help engineer even more accurate ANN models of biological vision, which in turn could feed back into machine vision.
“One thing is clear already: Improved ANN models … have led to control of a high-level neural population that was previously out of reach,” the authors said. “The results presented here have likely only scratched the surface of what is possible with such implemented characterizations of the brain’s neural networks.”
To Afraz, the power of AI here is to find cracks in human perception—both our computational models of sensory processes, as well as our evolved biological software itself. AI can be used “as a perfect adversarial tool to discover design cracks” of IT, said Afraz, such as finding computer art that “fools” a neuron into thinking the object is something else.
“As artificial intelligence researchers develop models that work as well as the brain does—or even better—we will still need to understand which networks are more likely to behave safely and further human goals,” said Ponce. “More efficient AI can be grounded by knowledge of how the brain works.”
Image Credit: Sangoiri / Shutterstock.com Continue reading
Scarcely a day goes by without another headline about neural networks: some new task that deep learning algorithms can excel at, approaching or even surpassing human competence. As the application of this approach to computer vision has continued to improve, with algorithms capable of specialized recognition tasks like those found in medicine, the software is getting closer to widespread commercial use—for example, in self-driving cars. Our ability to recognize patterns is a huge part of human intelligence: if this can be done faster by machines, the consequences will be profound.
Yet, as ever with algorithms, there are deep concerns about their reliability, especially when we don’t know precisely how they work. State-of-the-art neural networks will confidently—and incorrectly—classify images that look like television static or abstract art as real-world objects like school-buses or armadillos. Specific algorithms could be targeted by “adversarial examples,” where adding an imperceptible amount of noise to an image can cause an algorithm to completely mistake one object for another. Machine learning experts enjoy constructing these images to trick advanced software, but if a self-driving car could be fooled by a few stickers, it might not be so fun for the passengers.
These difficulties are hard to smooth out in large part because we don’t have a great intuition for how these neural networks “see” and “recognize” objects. The main insight analyzing a trained network itself can give us is a series of statistical weights, associating certain groups of points with certain objects: this can be very difficult to interpret.
Now, new research from UCLA, published in the journal PLOS Computational Biology, is testing neural networks to understand the limits of their vision and the differences between computer vision and human vision. Nicholas Baker, Hongjing Lu, and Philip J. Kellman of UCLA, alongside Gennady Erlikhman of the University of Nevada, tested a deep convolutional neural network called VGG-19. This is state-of-the-art technology that is already outperforming humans on standardized tests like the ImageNet Large Scale Visual Recognition Challenge.
They found that, while humans tend to classify objects based on their overall (global) shape, deep neural networks are far more sensitive to the textures of objects, including local color gradients and the distribution of points on the object. This result helps explain why neural networks in image recognition make mistakes that no human ever would—and could allow for better designs in the future.
In the first experiment, a neural network was trained to sort images into 1 of 1,000 different categories. It was then presented with silhouettes of these images: all of the local information was lost, while only the outline of the object remained. Ordinarily, the trained neural net was capable of recognizing these objects, assigning more than 90% probability to the correct classification. Studying silhouettes, this dropped to 10%. While human observers could nearly always produce correct shape labels, the neural networks appeared almost insensitive to the overall shape of the images. On average, the correct object was ranked as the 209th most likely solution by the neural network, even though the overall shapes were an exact match.
A particularly striking example arose when they tried to get the neural networks to classify glass figurines of objects they could already recognize. While you or I might find it easy to identify a glass model of an otter or a polar bear, the neural network classified them as “oxygen mask” and “can opener” respectively. By presenting glass figurines, where the texture information that neural networks relied on for classifying objects is lost, the neural network was unable to recognize the objects by shape alone. The neural network was similarly hopeless at classifying objects based on drawings of their outline.
If you got one of these right, you’re better than state-of-the-art image recognition software. Image Credit: Nicholas Baker, Hongjing Lu, Gennady Erlikhman, Philip J. Kelman. “Deep convolutional networks do not classify based on global object shape.” Plos Computational Biology. 12/7/18. / CC BY 4.0
When the neural network was explicitly trained to recognize object silhouettes—given no information in the training data aside from the object outlines—the researchers found that slight distortions or “ripples” to the contour of the image were again enough to fool the AI, while humans paid them no mind.
The fact that neural networks seem to be insensitive to the overall shape of an object—relying instead on statistical similarities between local distributions of points—suggests a further experiment. What if you scrambled the images so that the overall shape was lost but local features were preserved? It turns out that the neural networks are far better and faster at recognizing scrambled versions of objects than outlines, even when humans struggle. Students could classify only 37% of the scrambled objects, while the neural network succeeded 83% of the time.
Humans vastly outperform machines at classifying object (a) as a bear, while the machine learning algorithm has few problems classifying the bear in figure (b). Image Credit: Nicholas Baker, Hongjing Lu, Gennady Erlikhman, Philip J. Kelman. “Deep convolutional networks do not classify based on global object shape.” Plos Computational Biology. 12/7/18. / CC BY 4.0
“This study shows these systems get the right answer in the images they were trained on without considering shape,” Kellman said. “For humans, overall shape is primary for object recognition, and identifying images by overall shape doesn’t seem to be in these deep learning systems at all.”
Naively, one might expect that—as the many layers of a neural network are modeled on connections between neurons in the brain and resemble the visual cortex specifically—the way computer vision operates must necessarily be similar to human vision. But this kind of research shows that, while the fundamental architecture might resemble that of the human brain, the resulting “mind” operates very differently.
Researchers can, increasingly, observe how the “neurons” in neural networks light up when exposed to stimuli and compare it to how biological systems respond to the same stimuli. Perhaps someday it might be possible to use these comparisons to understand how neural networks are “thinking” and how those responses differ from humans.
But, as yet, it takes a more experimental psychology to probe how neural networks and artificial intelligence algorithms perceive the world. The tests employed against the neural network are closer to how scientists might try to understand the senses of an animal or the developing brain of a young child rather than a piece of software.
By combining this experimental psychology with new neural network designs or error-correction techniques, it may be possible to make them even more reliable. Yet this research illustrates just how much we still don’t understand about the algorithms we’re creating and using: how they tick, how they make decisions, and how they’re different from us. As they play an ever-greater role in society, understanding the psychology of neural networks will be crucial if we want to use them wisely and effectively—and not end up missing the woods for the trees.
Image Credit: Irvan Pratama / Shutterstock.com Continue reading
Every year, for just a few days in a major city, a small team of roboticists get to live the dream: ordering around their own personal robot butlers. In carefully-constructed replicas of a restaurant scene or a domestic setting, these robots perform any number of simple algorithmic tasks. “Get the can of beans from the shelf. Greet the visitors to the museum. Help the humans with their shopping. Serve the customers at the restaurant.”
This is Robocup @ Home, the annual tournament where teams of roboticists put their autonomous service robots to the test for practical domestic applications. The tasks seem simple and mundane, but considering the technology required reveals that they’re really not.
The Robot Butler Contest
Say you want a robot to fetch items in the supermarket. In a crowded, noisy environment, the robot must understand your commands, ask for clarification, and map out and navigate an unfamiliar environment, avoiding obstacles and people as it does so. Then it must recognize the product you requested, perhaps in a cluttered environment, perhaps in an unfamiliar orientation. It has to grasp that product appropriately—recall that there are entire multi-million-dollar competitions just dedicated to developing robots that can grasp a range of objects—and then return it to you.
It’s a job so simple that a child could do it—and so complex that teams of smart roboticists can spend weeks programming and engineering, and still end up struggling to complete simplified versions of this task. Of course, the child has the advantage of millions of years of evolutionary research and development, while the first robots that could even begin these tasks were only developed in the 1970s.
Even bearing this in mind, Robocup @ Home can feel like a place where futurist expectations come crashing into technologist reality. You dream of a smooth-voiced, sardonic JARVIS who’s already made your favorite dinner when you come home late from work; you end up shouting “remember the biscuits” at a baffled, ungainly droid in aisle five.
Caring for the Elderly
Famously, Japan is one of the most robo-enthusiastic nations in the world; they are the nation that stunned us all with ASIMO in 2000, and several studies have been conducted into the phenomenon. It’s no surprise, then, that humanoid robotics should be seriously considered as a solution to the crisis of the aging population. The Japanese government, as part of its robots strategy, has already invested $44 million in their development.
Toyota’s Human Support Robot (HSR-2) is a simple but programmable robot with a single arm; it can be remote-controlled to pick up objects and can monitor patients. HSR-2 has become the default robot for use in Robocup @ Home tournaments, at least in tasks that involve manipulating objects.
Alongside this, Toyota is working on exoskeletons to assist people in walking after strokes. It may surprise you to learn that nurses suffer back injuries more than any other occupation, at roughly three times the rate of construction workers, due to the day-to-day work of lifting patients. Toyota has a Care Assist robot/exoskeleton designed to fix precisely this problem by helping care workers with the heavy lifting.
The Home of the Future
The enthusiasm for domestic robotics is easy to understand and, in fact, many startups already sell robots marketed as domestic helpers in some form or another. In general, though, they skirt the immensely complicated task of building a fully capable humanoid robot—a task that even Google’s skunk-works department gave up on, at least until recently.
It’s plain to see why: far more research and development is needed before these domestic robots could be used reliably and at a reasonable price. Consumers with expectations inflated by years of science fiction saturation might find themselves frustrated as the robots fail to perform basic tasks.
Instead, domestic robotics efforts fall into one of two categories. There are robots specialized to perform a domestic task, like iRobot’s Roomba, which stuck to vacuuming and became the most successful domestic robot of all time by far.
The tasks need not necessarily be simple, either: the impressive but expensive automated kitchen uses the world’s most dexterous hands to cook meals, providing it can recognize the ingredients. Other robots focus on human-robot interaction, like Jibo: they essentially package the abilities of a voice assistant like Siri, Cortana, or Alexa to respond to simple questions and perform online tasks in a friendly, dynamic robot exterior.
In this way, the future of domestic automation starts to look a lot more like smart homes than a robot or domestic servant. General robotics is difficult in the same way that general artificial intelligence is difficult; competing with humans, the great all-rounders, is a challenge. Getting superhuman performance at a more specific task, however, is feasible and won’t cost the earth.
Individual startups without the financial might of a Google or an Amazon can develop specialized robots, like Seven Dreamers’ laundry robot, and hope that one day it will form part of a network of autonomous robots that each have a role to play in the household.
The Smart Home has been a staple of futurist expectations for a long time, to the extent that movies featuring smart homes out of control are already a cliché. But critics of the smart home idea—and of the internet of things more generally—tend to focus on the idea that, more often than not, software just adds an additional layer of things that can break (NSFW), in exchange for minimal added convenience. A toaster that can short-circuit is bad enough, but a toaster that can refuse to serve you toast because its firmware is updating is something else entirely.
That’s before you even get into the security vulnerabilities, which are all the more important when devices are installed in your home and capable of interacting with them. The idea of a smart watch that lets you keep an eye on your children might sound like something a security-conscious parent would like: a smart watch that can be hacked to track children, listen in on their surroundings, and even fool them into thinking a call is coming from their parents is the stuff of nightmares.
Key to many of these problems is the lack of standardization for security protocols, and even the products themselves. The idea of dozens of startups each developing a highly-specialized piece of robotics to perform a single domestic task sounds great in theory, until you realize the potential hazards and pitfalls of getting dozens of incompatible devices to work together on the same system.
It seems inevitable that there are yet more layers of domestic drudgery that can be automated away, decades after the first generation of time-saving domestic devices like the dishwasher and vacuum cleaner became mainstream. With projected market values into the billions and trillions of dollars, there is no shortage of industry interest in ironing out these kinks. But, for now at least, the answer to the question: “Where’s my robot butler?” is that it is gradually, painstakingly learning how to sort through groceries.
Image Credit: Nonchanon / Shutterstock.com Continue reading
Autonomous vehicles can follow the general rules of American roads, recognizing traffic signals and lane markings, noticing crosswalks and other regular features of the streets. But they work only on well-marked roads that are carefully scanned and mapped in advance.
Many paved roads, though, have faded paint, signs obscured behind trees and unusual intersections. In addition, 1.4 million miles of U.S. roads—one-third of the country’s public roadways—are unpaved, with no on-road signals like lane markings or stop-here lines. That doesn’t include miles of private roads, unpaved driveways or off-road trails.
What’s a rule-following autonomous car to do when the rules are unclear or nonexistent? And what are its passengers to do when they discover their vehicle can’t get them where they’re going?
Accounting for the Obscure
Most challenges in developing advanced technologies involve handling infrequent or uncommon situations, or events that require performance beyond a system’s normal capabilities. That’s definitely true for autonomous vehicles. Some on-road examples might be navigating construction zones, encountering a horse and buggy, or seeing graffiti that looks like a stop sign. Off-road, the possibilities include the full variety of the natural world, such as trees down over the road, flooding and large puddles—or even animals blocking the way.
At Mississippi State University’s Center for Advanced Vehicular Systems, we have taken up the challenge of training algorithms to respond to circumstances that almost never happen, are difficult to predict and are complex to create. We seek to put autonomous cars in the hardest possible scenario: driving in an area the car has no prior knowledge of, with no reliable infrastructure like road paint and traffic signs, and in an unknown environment where it’s just as likely to see a cactus as a polar bear.
Our work combines virtual technology and the real world. We create advanced simulations of lifelike outdoor scenes, which we use to train artificial intelligence algorithms to take a camera feed and classify what it sees, labeling trees, sky, open paths and potential obstacles. Then we transfer those algorithms to a purpose-built all-wheel-drive test vehicle and send it out on our dedicated off-road test track, where we can see how our algorithms work and collect more data to feed into our simulations.
We have developed a simulator that can create a wide range of realistic outdoor scenes for vehicles to navigate through. The system generates a range of landscapes of different climates, like forests and deserts, and can show how plants, shrubs and trees grow over time. It can also simulate weather changes, sunlight and moonlight, and the accurate locations of 9,000 stars.
The system also simulates the readings of sensors commonly used in autonomous vehicles, such as lidar and cameras. Those virtual sensors collect data that feeds into neural networks as valuable training data.
Simulated desert, meadow and forest environments generated by the Mississippi State University Autonomous Vehicle Simulator. Chris Goodin, Mississippi State University, Author provided.
Building a Test Track
Simulations are only as good as their portrayals of the real world. Mississippi State University has purchased 50 acres of land on which we are developing a test track for off-road autonomous vehicles. The property is excellent for off-road testing, with unusually steep grades for our area of Mississippi—up to 60 percent inclines—and a very diverse population of plants.
We have selected certain natural features of this land that we expect will be particularly challenging for self-driving vehicles, and replicated them exactly in our simulator. That allows us to directly compare results from the simulation and real-life attempts to navigate the actual land. Eventually, we’ll create similar real and virtual pairings of other types of landscapes to improve our vehicle’s capabilities.
A road washout, as seen in real life, left, and in simulation. Chris Goodin, Mississippi State University, Author provided.
Collecting More Data
We have also built a test vehicle, called the Halo Project, which has an electric motor and sensors and computers that can navigate various off-road environments. The Halo Project car has additional sensors to collect detailed data about its actual surroundings, which can help us build virtual environments to run new tests in.
The Halo Project car can collect data about driving and navigating in rugged terrain. Beth Newman Wynn, Mississippi State University, Author provided.
Two of its lidar sensors, for example, are mounted at intersecting angles on the front of the car so their beams sweep across the approaching ground. Together, they can provide information on how rough or smooth the surface is, as well as capturing readings from grass and other plants and items on the ground.
Lidar beams intersect, scanning the ground in front of the vehicle. Chris Goodin, Mississippi State University, Author provided
We’ve seen some exciting early results from our research. For example, we have shown promising preliminary results that machine learning algorithms trained on simulated environments can be useful in the real world. As with most autonomous vehicle research, there is still a long way to go, but our hope is that the technologies we’re developing for extreme cases will also help make autonomous vehicles more functional on today’s roads.
Matthew Doude, Associate Director, Center for Advanced Vehicular Systems; Ph.D. Student in Industrial and Systems Engineering, Mississippi State University; Christopher Goodin, Assistant Research Professor, Center for Advanced Vehicular Systems, Mississippi State University, and Daniel Carruth, Assistant Research Professor and Associate Director for Human Factors and Advanced Vehicle System, Center for Advanced Vehicular Systems, Mississippi State University
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Photo provided for The Conversation by Matthew Goudin / CC BY ND Continue reading