Tag Archives: institute

#435213 Robot traps ball without coding

Dr. Kee-hoon Kim's team at the Center for Intelligent & Interactive Robotics of the Korea Institute of Science and Technology (KIST) developed a way of teaching “impedance-controlled robots” through human demonstrations using surface electromyograms (sEMG) of muscles, and succeeded in teaching a robot to trap a dropped ball like a soccer player. A surface electromyogram is an electric signal produced during muscle activation that can be picked up on the surface of the skin. Continue reading

Posted in Human Robots

#435152 The Futuristic Tech Disrupting Real ...

In the wake of the housing market collapse of 2008, one entrepreneur decided to dive right into the failing real estate industry. But this time, he didn’t buy any real estate to begin with. Instead, Glenn Sanford decided to launch the first-ever cloud-based real estate brokerage, eXp Realty.

Contracting virtual platform VirBELA to build out the company’s mega-campus in VR, eXp Realty demonstrates the power of a dematerialized workspace, throwing out hefty overhead costs and fundamentally redefining what ‘real estate’ really means. Ten years later, eXp Realty has an army of 14,000 agents across all 50 US states, 3 Canadian provinces, and 400 MLS market areas… all without a single physical office.

But VR is just one of many exponential technologies converging to revolutionize real estate and construction. As floating cities and driverless cars spread out your living options, AI and VR are together cutting out the middleman.

Already, the global construction industry is projected to surpass $12.9 trillion in 2022, and the total value of the US housing market alone grew to $33.3 trillion last year. Both vital for our daily lives, these industries will continue to explode in value, posing countless possibilities for disruption.

In this blog, I’ll be discussing the following trends:

New prime real estate locations;
Disintermediation of the real estate broker and search;
Materials science and 3D printing in construction.

Let’s dive in!

Location Location Location
Until today, location has been the name of the game when it comes to hunting down the best real estate. But constraints on land often drive up costs while limiting options, and urbanization is only exacerbating the problem.

Beyond the world of virtual real estate, two primary mechanisms are driving the creation of new locations.

(1) Floating Cities

Offshore habitation hubs, floating cities have long been conceived as a solution to rising sea levels, skyrocketing urban populations, and threatened ecosystems. In success, they will soon unlock an abundance of prime real estate, whether for scenic living, commerce, education, or recreation.

One pioneering model is that of Oceanix City, designed by Danish architect Bjarke Ingels and a host of other domain experts. Intended to adapt organically over time, Oceanix would consist of a galaxy of mass-produced, hexagonal floating modules, built as satellite “cities” off coastal urban centers and sustained by renewable energies.

While individual 4.5-acre platforms would each sustain 300 people, these hexagonal modules are designed to link into 75-acre tessellations sustaining up to 10,000 residents. Each anchored to the ocean floor using biorock, Oceanix cities are slated to be closed-loop systems, as external resources are continuously supplied by automated drone networks.

Electric boats or flying cars might zoom you to work, city-embedded water capture technologies would provide your water, and while vertical and outdoor farming supply your family meal, share economies would dominate goods provision.

AERIAL: Located in calm, sheltered waters, near coastal megacities, OCEANIX City will be an adaptable, sustainable, scalable, and affordable solution for human life on the ocean. Image Credit: OCEANIX/BIG-Bjarke Ingels Group.
Joined by countless government officials whose islands risk submersion at the hands of sea level rise, the UN is now getting on board. And just this year, seasteading is exiting the realm of science fiction and testing practical waters.

As French Polynesia seeks out robust solutions to sea level rise, their government has now joined forces with the San Francisco-based Seasteading Institute. With a newly designated special economic zone and 100 acres of beachfront, this joint Floating Island Project could even see up to a dozen inhabitable structures by 2020. And what better to fund the $60 million project than the team’s upcoming ICO?

But aside from creating new locations, autonomous vehicles (AVs) and flying cars are turning previously low-demand land into the prime real estate of tomorrow.

(2) Autonomous Electric Vehicles and Flying Cars

Today, the value of a location is a function of its proximity to your workplace, your city’s central business district, the best schools, or your closest friends.

But what happens when driverless cars desensitize you to distance, or Hyperloop and flying cars decimate your commute time? Historically, every time new transit methods have hit the mainstream, tolerance for distance has opened up right alongside them, further catalyzing city spread.

And just as Hyperloop and the Boring Company aim to make your commute immaterial, autonomous vehicle (AV) ridesharing services will spread out cities in two ways: (1) by drastically reducing parking spaces needed (vertical parking decks = more prime real estate); and (2) by untethering you from the steering wheel. Want an extra two hours of sleep on the way to work? Schedule a sleeper AV and nap on your route to the office. Need a car-turned-mobile-office? No problem.

Meanwhile, aerial taxis (i.e. flying cars) will allow you to escape ground congestion entirely, delivering you from bedroom to boardroom at decimated time scales.

Already working with regulators, Uber Elevate has staked ambitious plans for its UberAIR airborne taxi project. By 2023, Uber anticipates rolling out flying drones in its two first pilot cities, Los Angeles and Dallas. Flying between rooftop skyports, drones would carry passengers at a height of 1,000 to 2,000 feet at speeds between 100 to 200 mph. And while costs per ride are anticipated to resemble those of an Uber Black based on mileage, prices are projected to soon drop to those of an UberX.

But the true economic feat boils down to this: if I were to commute 50 to 100 kilometers, I could get two or three times the house for the same price. (Not to mention the extra living space offered up by my now-unneeded garage.)

All of a sudden, virtual reality, broadband, AVs, or high-speed vehicles are going to change where we live and where we work. So rather than living in a crowded, dense urban core for access to jobs and entertainment, our future of personalized, autonomous, low-cost transport opens the luxury of rural areas to all without compromising the benefits of a short commute.

Once these drivers multiply your real estate options, how will you select your next home?

Disintermediation: Say Bye to Your Broker
In a future of continuous and personalized preference-tracking, why hire a human agent who knows less about your needs and desires than a personal AI?

Just as disintermediation is cutting out bankers and insurance agents, so too is it closing in on real estate brokers. Over the next decade, as AI becomes your agent, VR will serve as your medium.

To paint a more vivid picture of how this will look, over 98 percent of your home search will be conducted from the comfort of your couch through next-generation VR headgear.

Once you’ve verbalized your primary desires for home location, finishings, size, etc. to your personal AI, it will offer you top picks, tour-able 24/7, with optional assistance by a virtual guide and constantly updated data. As a seller, this means potential buyers from two miles, or two continents, away.

Throughout each immersive VR tour, advanced eye-tracking software and a permissioned machine learning algorithm follow your gaze, further learn your likes and dislikes, and intelligently recommend other homes or commercial residences to visit.

Curious as to what the living room might look like with a fresh coat of blue paint and a white carpet? No problem! VR programs will be able to modify rendered environments instantly, changing countless variables, from furniture materials to even the sun’s orientation. Keen to input your own furniture into a VR-rendered home? Advanced AIs could one day compile all your existing furniture, electronics, clothing, decorations, and even books, virtually organizing them across any accommodating new space.

As 3D scanning technologies make extraordinary headway, VR renditions will only grow cheaper and higher resolution. One company called Immersive Media (disclosure: I’m an investor and advisor) has a platform for 360-degree video capture and distribution, and is already exploring real estate 360-degree video.

Smaller firms like Studio 216, Vieweet, Arch Virtual, ArX Solutions, and Rubicon Media can similarly capture and render models of various properties for clients and investors to view and explore. In essence, VR real estate platforms will allow you to explore any home for sale, do the remodel, and determine if it truly is the house of your dreams.

Once you’re ready to make a bid, your AI will even help estimate a bid, process and submit your offer. Real estate companies like Zillow, Trulia, Move, Redfin, ZipRealty (acquired by Realogy in 2014) and many others have already invested millions in machine learning applications to make search, valuation, consulting, and property management easier, faster, and much more accurate.

But what happens if the home you desire most means starting from scratch with new construction?

New Methods and Materials for Construction
For thousands of years, we’ve been constrained by the construction materials of nature. We built bricks from naturally abundant clay and shale, used tree limbs as our rooftops and beams, and mastered incredible structures in ancient Rome with the use of cement.

But construction is now on the cusp of a materials science revolution. Today, I’d like to focus on three key materials:

Upcycled Materials

Imagine if you could turn the world’s greatest waste products into their most essential building blocks. Thanks to UCLA researchers at CO2NCRETE, we can already do this with carbon emissions.

Today, concrete produces about five percent of all greenhouse gas (GHG) emissions. But what if concrete could instead conserve greenhouse emissions? CO2NCRETE engineers capture carbon from smokestacks and combine it with lime to create a new type of cement. The lab’s 3D printers then shape the upcycled concrete to build entirely new structures. Once conquered at scale, upcycled concrete will turn a former polluter into a future conserver.

Or what if we wanted to print new residences from local soil at hand? Marking an extraordinary convergence between robotics and 3D printing, the Institute of Advanced Architecture of Catalonia (IAAC) is already working on a solution.

In a major feat for low-cost construction in remote zones, IAAC has found a way to convert almost any soil into a building material with three times the tensile strength of industrial clay. Offering myriad benefits, including natural insulation, low GHG emissions, fire protection, air circulation, and thermal mediation, IAAC’s new 3D printed native soil can build houses on-site for as little as $1,000.

Nanomaterials

Nano- and micro-materials are ushering in a new era of smart, super-strong, and self-charging buildings. While carbon nanotubes dramatically increase the strength-to-weight ratio of skyscrapers, revolutionizing their structural flexibility, nanomaterials don’t stop here.

Several research teams are pioneering silicon nanoparticles to capture everyday light flowing through our windows. Little solar cells at the edges of windows then harvest this energy for ready use. Researchers at the US National Renewable Energy Lab have developed similar smart windows. Turning into solar panels when bathed in sunlight, these thermochromic windows will power our buildings, changing color as they do.

Self-Healing Infrastructure

The American Society of Civil Engineers estimates that the US needs to spend roughly $4.5 trillion to fix nationwide roads, bridges, dams, and common infrastructure by 2025. But what if infrastructure could fix itself?

Enter self-healing concrete. Engineers at Delft University have developed bio-concrete that can repair its own cracks. As head researcher Henk Jonkers explains, “What makes this limestone-producing bacteria so special is that they are able to survive in concrete for more than 200 years and come into play when the concrete is damaged. […] If cracks appear as a result of pressure on the concrete, the concrete will heal these cracks itself.”

But bio-concrete is only the beginning of self-healing technologies. As futurist architecture firms start printing plastic and carbon-fiber houses like the stunner seen below (using Branch Technologies’ 3D printing technology), engineers have begun tackling self-healing plastic.

And in a bid to go smart, burgeoning construction projects have started embedding sensors for preemptive detection. Beyond materials and sensors, however, construction methods are fast colliding into robotics and 3D printing.

While some startups and research institutes have leveraged robot swarm construction (namely, Harvard’s robotic termite-like swarm of programmed constructors), others have taken to large-scale autonomous robots.

One such example involves Fastbrick Robotics. After multiple iterations, the company’s Hadrian X end-to-end bricklaying robot can now autonomously build a fully livable, 180-square meter home in under 3 days. Using a laser-guided robotic attachment, the all-in-one brick-loaded truck simply drives to a construction site and directs blocks through its robotic arm in accordance with a 3D model.

Layhead. Image Credit: Fastbrick Robotics.
Meeting verified building standards, Hadrian and similar solutions hold massive promise in the long term, deployable across post-conflict refugee sites and regions recovering from natural catastrophes.

Imagine the implications. Eliminating human safety concerns and unlocking any environment, autonomous builder robots could collaboratively build massive structures in space or deep underwater habitats.

Final Thoughts
Where, how, and what we live in form a vital pillar of our everyday lives. The concept of “home” is unlikely to disappear anytime soon. At the same time, real estate and construction are two of the biggest playgrounds for technological convergence, each on the verge of revolutionary disruption.

As underlying shifts in transportation, land reclamation, and the definition of “space” (real vs. virtual) take hold, the real estate market is about to explode in value, spreading out urban centers on unprecedented scales and unlocking vast new prime “property.”

Meanwhile, converging advancements in AI and VR are fundamentally disrupting the way we design, build, and explore new residences. Just as mirror worlds create immersive, virtual real estate economies, VR tours and AI agents are absorbing both sides of the coin to entirely obliterate the middleman.

And as materials science breakthroughs meet new modes of construction, the only limits to tomorrow’s structures are those of our own imagination.

Join Me
Abundance-Digital Online Community: Stay ahead of technological advancements and turn your passion into action. Abundance Digital is now part of Singularity University. Learn more.

Image Credit: OCEANIX/BIG-Bjarke Ingels Group. Continue reading

Posted in Human Robots

#435127 Teaching AI the Concept of ‘Similar, ...

As a human you instinctively know that a leopard is closer to a cat than a motorbike, but the way we train most AI makes them oblivious to these kinds of relations. Building the concept of similarity into our algorithms could make them far more capable, writes the author of a new paper in Science Robotics.

Convolutional neural networks have revolutionized the field of computer vision to the point that machines are now outperforming humans on some of the most challenging visual tasks. But the way we train them to analyze images is very different from the way humans learn, says Atsuto Maki, an associate professor at KTH Royal Institute of Technology.

“Imagine that you are two years old and being quizzed on what you see in a photo of a leopard,” he writes. “You might answer ‘a cat’ and your parents might say, ‘yeah, not quite but similar’.”

In contrast, the way we train neural networks rarely gives that kind of partial credit. They are typically trained to have very high confidence in the correct label and consider all incorrect labels, whether ”cat” or “motorbike,” equally wrong. That’s a mistake, says Maki, because ignoring the fact that something can be “less wrong” means you’re not exploiting all of the information in the training data.

Even when models are trained this way, there will be small differences in the probabilities assigned to incorrect labels that can tell you a lot about how well the model can generalize what it has learned to unseen data.

If you show a model a picture of a leopard and it gives “cat” a probability of five percent and “motorbike” one percent, that suggests it picked up on the fact that a cat is closer to a leopard than a motorbike. In contrast, if the figures are the other way around it means the model hasn’t learned the broad features that make cats and leopards similar, something that could potentially be helpful when analyzing new data.

If we could boost this ability to identify similarities between classes we should be able to create more flexible models better able to generalize, says Maki. And recent research has demonstrated how variations of an approach called regularization might help us achieve that goal.

Neural networks are prone to a problem called “overfitting,” which refers to a tendency to pay too much attention to tiny details and noise specific to their training set. When that happens, models will perform excellently on their training data but poorly when applied to unseen test data without these particular quirks.

Regularization is used to circumvent this problem, typically by reducing the network’s capacity to learn all this unnecessary information and therefore boost its ability to generalize to new data. Techniques are varied, but generally involve modifying the network’s structure or the strength of the weights between artificial neurons.

More recently, though, researchers have suggested new regularization approaches that work by encouraging a broader spread of probabilities across all classes. This essentially helps them capture more of the class similarities, says Maki, and therefore boosts their ability to generalize.

One such approach was devised in 2017 by Google Brain researchers, led by deep learning pioneer Geoffrey Hinton. They introduced a penalty to their training process that directly punished overconfident predictions in the model’s outputs, and a technique called label smoothing that prevents the largest probability becoming much larger than all others. This meant the probabilities were lower for correct labels and higher for incorrect ones, which was found to boost performance of models on varied tasks from image classification to speech recognition.

Another came from Maki himself in 2017 and achieves the same goal, but by suppressing high values in the model’s feature vector—the mathematical construct that describes all of an object’s important characteristics. This has a knock-on effect on the spread of output probabilities and also helped boost performance on various image classification tasks.

While it’s still early days for the approach, the fact that humans are able to exploit these kinds of similarities to learn more efficiently suggests that models that incorporate them hold promise. Maki points out that it could be particularly useful in applications such as robotic grasping, where distinguishing various similar objects is important.

Image Credit: Marianna Kalashnyk / Shutterstock.com Continue reading

Posted in Human Robots

#435098 Coming of Age in the Age of AI: The ...

The first generation to grow up entirely in the 21st century will never remember a time before smartphones or smart assistants. They will likely be the first children to ride in self-driving cars, as well as the first whose healthcare and education could be increasingly turned over to artificially intelligent machines.

Futurists, demographers, and marketers have yet to agree on the specifics of what defines the next wave of humanity to follow Generation Z. That hasn’t stopped some, like Australian futurist Mark McCrindle, from coining the term Generation Alpha, denoting a sort of reboot of society in a fully-realized digital age.

“In the past, the individual had no power, really,” McCrindle told Business Insider. “Now, the individual has great control of their lives through being able to leverage this world. Technology, in a sense, transformed the expectations of our interactions.”

No doubt technology may impart Marvel superhero-like powers to Generation Alpha that even tech-savvy Millennials never envisioned over cups of chai latte. But the powers of machine learning, computer vision, and other disciplines under the broad category of artificial intelligence will shape this yet unformed generation more definitively than any before it.

What will it be like to come of age in the Age of AI?

The AI Doctor Will See You Now
Perhaps no other industry is adopting and using AI as much as healthcare. The term “artificial intelligence” appears in nearly 90,000 publications from biomedical literature and research on the PubMed database.

AI is already transforming healthcare and longevity research. Machines are helping to design drugs faster and detect disease earlier. And AI may soon influence not only how we diagnose and treat illness in children, but perhaps how we choose which children will be born in the first place.

A study published earlier this month in NPJ Digital Medicine by scientists from Weill Cornell Medicine used 12,000 photos of human embryos taken five days after fertilization to train an AI algorithm on how to tell which in vitro fertilized embryo had the best chance of a successful pregnancy based on its quality.

Investigators assigned each embryo a grade based on various aspects of its appearance. A statistical analysis then correlated that grade with the probability of success. The algorithm, dubbed Stork, was able to classify the quality of a new set of images with 97 percent accuracy.

“Our algorithm will help embryologists maximize the chances that their patients will have a single healthy pregnancy,” said Dr. Olivier Elemento, director of the Caryl and Israel Englander Institute for Precision Medicine at Weill Cornell Medicine, in a press release. “The IVF procedure will remain the same, but we’ll be able to improve outcomes by harnessing the power of artificial intelligence.”

Other medical researchers see potential in applying AI to detect possible developmental issues in newborns. Scientists in Europe, working with a Finnish AI startup that creates seizure monitoring technology, have developed a technique for detecting movement patterns that might indicate conditions like cerebral palsy.

Published last month in the journal Acta Pediatrica, the study relied on an algorithm to extract the movements from a newborn, turning it into a simplified “stick figure” that medical experts could use to more easily detect clinically relevant data.

The researchers are continuing to improve the datasets, including using 3D video recordings, and are now developing an AI-based method for determining if a child’s motor maturity aligns with its true age. Meanwhile, a study published in February in Nature Medicine discussed the potential of using AI to diagnose pediatric disease.

AI Gets Classy
After being weaned on algorithms, Generation Alpha will hit the books—about machine learning.

China is famously trying to win the proverbial AI arms race by spending billions on new technologies, with one Chinese city alone pledging nearly $16 billion to build a smart economy based on artificial intelligence.

To reach dominance by its stated goal of 2030, Chinese cities are also incorporating AI education into their school curriculum. Last year, China published its first high school textbook on AI, according to the South China Morning Post. More than 40 schools are participating in a pilot program that involves SenseTime, one of the country’s biggest AI companies.

In the US, where it seems every child has access to their own AI assistant, researchers are just beginning to understand how the ubiquity of intelligent machines will influence the ways children learn and interact with their highly digitized environments.

Sandra Chang-Kredl, associate professor of the department of education at Concordia University, told The Globe and Mail that AI could have detrimental effects on learning creativity or emotional connectedness.

Similar concerns inspired Stefania Druga, a member of the Personal Robots group at the MIT Media Lab (and former Education Teaching Fellow at SU), to study interactions between children and artificial intelligence devices in order to encourage positive interactions.

Toward that goal, Druga created Cognimates, a platform that enables children to program and customize their own smart devices such as Alexa or even a smart, functional robot. The kids can also use Cognimates to train their own AI models or even build a machine learning version of Rock Paper Scissors that gets better over time.

“I believe it’s important to also introduce young people to the concepts of AI and machine learning through hands-on projects so they can make more informed and critical use of these technologies,” Druga wrote in a Medium blog post.

Druga is also the founder of Hackidemia, an international organization that sponsors workshops and labs around the world to introduce kids to emerging technologies at an early age.

“I think we are in an arms race in education with the advancement of technology, and we need to start thinking about AI literacy before patterns of behaviors for children and their families settle in place,” she wrote.

AI Goes Back to School
It also turns out that AI has as much to learn from kids. More and more researchers are interested in understanding how children grasp basic concepts that still elude the most advanced machine minds.

For example, developmental psychologist Alison Gopnik has written and lectured extensively about how studying the minds of children can provide computer scientists clues on how to improve machine learning techniques.

In an interview on Vox, she described that while DeepMind’s AlpahZero was trained to be a chessmaster, it struggles with even the simplest changes in the rules, such as allowing the bishop to move horizontally instead of vertically.

“A human chess player, even a kid, will immediately understand how to transfer that new rule to their playing of the game,” she noted. “Flexibility and generalization are something that even human one-year-olds can do but that the best machine learning systems have a much harder time with.”

Last year, the federal defense agency DARPA announced a new program aimed at improving AI by teaching it “common sense.” One of the chief strategies is to develop systems for “teaching machines through experience, mimicking the way babies grow to understand the world.”

Such an approach is also the basis of a new AI program at MIT called the MIT Quest for Intelligence.

The research leverages cognitive science to understand human intelligence, according to an article on the project in MIT Technology Review, such as exploring how young children visualize the world using their own innate 3D models.

“Children’s play is really serious business,” said Josh Tenenbaum, who leads the Computational Cognitive Science lab at MIT and his head of the new program. “They’re experiments. And that’s what makes humans the smartest learners in the known universe.”

In a world increasingly driven by smart technologies, it’s good to know the next generation will be able to keep up.

Image Credit: phoelixDE / Shutterstock.com Continue reading

Posted in Human Robots

#435056 How Researchers Used AI to Better ...

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.

Why Vision?
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

Posted in Human Robots