Tag Archives: chinese

#437120 The New Indiana Jones? AI. Here’s How ...

Archaeologists have uncovered scores of long-abandoned settlements along coastal Madagascar that reveal environmental connections to modern-day communities. They have detected the nearly indiscernible bumps of earthen mounds left behind by prehistoric North American cultures. Still other researchers have mapped Bronze Age river systems in the Indus Valley, one of the cradles of civilization.

All of these recent discoveries are examples of landscape archaeology. They’re also examples of how artificial intelligence is helping scientists hunt for new archaeological digs on a scale and at a pace unimaginable even a decade ago.

“AI in archaeology has been increasing substantially over the past few years,” said Dylan Davis, a PhD candidate in the Department of Anthropology at Penn State University. “One of the major uses of AI in archaeology is for the detection of new archaeological sites.”

The near-ubiquitous availability of satellite data and other types of aerial imagery for many parts of the world has been both a boon and a bane to archaeologists. They can cover far more ground, but the job of manually mowing their way across digitized landscapes is still time-consuming and laborious. Machine learning algorithms offer a way to parse through complex data far more quickly.

AI Gives Archaeologists a Bird’s Eye View
Davis developed an automated algorithm for identifying large earthen and shell mounds built by native populations long before Europeans arrived with far-off visions of skyscrapers and superhighways in their eyes. The sites still hidden in places like the South Carolina wilderness contain a wealth of information about how people lived, even what they ate, and the ways they interacted with the local environment and other cultures.

In this particular case, the imagery comes from LiDAR, which uses light pulses that can penetrate tree canopies to map forest floors. The team taught the computer the shape, size, and texture characteristics of the mounds so it could identify potential sites from the digital 3D datasets that it analyzed.

“The process resulted in several thousand possible features that my colleagues and I checked by hand,” Davis told Singularity Hub. “While not entirely automated, this saved the equivalent of years of manual labor that would have been required for analyzing the whole LiDAR image by hand.”

In Madagascar—where Davis is studying human settlement history across the world’s fourth largest island over a timescale of millennia—he developed a predictive algorithm to help locate archaeological sites using freely available satellite imagery. His team was able to survey and identify more than 70 new archaeological sites—and potentially hundreds more—across an area of more than 1,000 square kilometers during the course of about a year.

Machines Learning From the Past Prepare Us for the Future
One impetus behind the rapid identification of archaeological sites is that many are under threat from climate change, such as coastal erosion from sea level rise, or other human impacts. Meanwhile, traditional archaeological approaches are expensive and laborious—serious handicaps in a race against time.

“It is imperative to record as many archaeological sites as we can in a short period of time. That is why AI and machine learning are useful for my research,” Davis said.

Studying the rise and fall of past civilizations can also teach modern humans a thing or two about how to grapple with these current challenges.

Researchers at the Institut Català d’Arqueologia Clàssica (ICAC) turned to machine-learning algorithms to reconstruct more than 20,000 kilometers of paleo-rivers along the Indus Valley civilization of what is now part of modern Pakistan and India. Such AI-powered mapping techniques wouldn’t be possible using satellite images alone.

That effort helped locate many previously unknown archaeological sites and unlocked new insights into those Bronze Age cultures. However, the analytics can also assist governments with important water resource management today, according to Hèctor A. Orengo Romeu, co-director of the Landscape Archaeology Research Group at ICAC.

“Our analyses can contribute to the forecasts of the evolution of aquifers in the area and provide valuable information on aspects such as the variability of agricultural productivity or the influence of climate change on the expansion of the Thar desert, in addition to providing cultural management tools to the government,” he said.

Leveraging AI for Language and Lots More
While landscape archaeology is one major application of AI in archaeology, it’s far from the only one. In 2000, only about a half-dozen scientific papers referred to the use of AI, according to the Web of Science, reputedly the world’s largest global citation database. Last year, more than 65 papers were published concerning the use of machine intelligence technologies in archaeology, with a significant uptick beginning in 2015.

AI methods, for instance, are being used to understand the chemical makeup of artifacts like pottery and ceramics, according to Davis. “This can help identify where these materials were made and how far they were transported. It can also help us to understand the extent of past trading networks.”

Linguistic anthropologists have also used machine intelligence methods to trace the evolution of different languages, Davis said. “Using AI, we can learn when and where languages emerged around the world.”

In other cases, AI has helped reconstruct or decipher ancient texts. Last year, researchers at Google’s DeepMind used a deep neural network called PYTHIA to recreate missing inscriptions in ancient Greek from damaged surfaces of objects made of stone or ceramics.

Named after the Oracle at Delphi, PYTHIA “takes a sequence of damaged text as input, and is trained to predict character sequences comprising hypothesised restorations of ancient Greek inscriptions,” the researchers reported.

In a similar fashion, Chinese scientists applied a convolutional neural network (CNN) to untangle another ancient tongue once found on turtle shells and ox bones. The CNN managed to classify oracle bone morphology in order to piece together fragments of these divination objects, some with inscriptions that represent the earliest evidence of China’s recorded history.

“Differentiating the materials of oracle bones is one of the most basic steps for oracle bone morphology—we need to first make sure we don’t assemble pieces of ox bones with tortoise shells,” lead author of the study, associate professor Shanxiong Chen at China’s Southwest University, told Synced, an online tech publication in China.

AI Helps Archaeologists Get the Scoop…
And then there are applications of AI in archaeology that are simply … interesting. Just last month, researchers published a paper about a machine learning method trained to differentiate between human and canine paleofeces.

The algorithm, dubbed CoproID, compares the gut microbiome DNA found in the ancient material with DNA found in modern feces, enabling it to get the scoop on the origin of the poop.

Also known as coprolites, paleo-feces from humans and dogs are often found in the same archaeological sites. Scientists need to know which is which if they’re trying to understand something like past diets or disease.

“CoproID is the first line of identification in coprolite analysis to confirm that what we’re looking for is actually human, or a dog if we’re interested in dogs,” Maxime Borry, a bioinformatics PhD student at the Max Planck Institute for the Science of Human History, told Vice.

…But Machine Intelligence Is Just Another Tool
There is obviously quite a bit of work that can be automated through AI. But there’s no reason for archaeologists to hit the unemployment line any time soon. There are also plenty of instances where machines can’t yet match humans in identifying objects or patterns. At other times, it’s just faster doing the analysis yourself, Davis noted.

“For ‘big data’ tasks like detecting archaeological materials over a continental scale, AI is useful,” he said. “But for some tasks, it is sometimes more time-consuming to train an entire computer algorithm to complete a task that you can do on your own in an hour.”

Still, there’s no telling what the future will hold for studying the past using artificial intelligence.

“We have already started to see real improvements in the accuracy and reliability of these approaches, but there is a lot more to do,” Davis said. “Hopefully, we start to see these methods being directly applied to a variety of interesting questions around the world, as these methods can produce datasets that would have been impossible a few decades ago.”

Image Credit: James Wheeler from Pixabay Continue reading

Posted in Human Robots

#436546 How AI Helped Predict the Coronavirus ...

Coronavirus has been all over the news for the last couple weeks. A dedicated hospital sprang up in just eight days, the stock market took a hit, Chinese New Year celebrations were spoiled, and travel restrictions are in effect.

But let’s rewind a bit; some crucial events took place before we got to this point.

A little under two weeks before the World Health Organization (WHO) alerted the public of the coronavirus outbreak, a Canadian artificial intelligence company was already sounding the alarm. BlueDot uses AI-powered algorithms to analyze information from a multitude of sources to identify disease outbreaks and forecast how they may spread. On December 31st 2019, the company sent out a warning to its customers to avoid Wuhan, where the virus originated. The WHO didn’t send out a similar public notice until January 9th, 2020.

The story of BlueDot’s early warning is the latest example of how AI can improve our identification of and response to new virus outbreaks.

Predictions Are Bad News
Global pandemic or relatively minor scare? The jury is still out on the coronavirus. However, the math points to signs that the worst is yet to come.

Scientists are still working to determine how infectious the virus is. Initial analysis suggests it may be somewhere between influenza and polio on the virus reproduction number scale, which indicates how many new cases one case leads to.

UK and US-based researchers have published a preliminary paper estimating that the confirmed infected people in Wuhan only represent five percent of those who are actually infected. If the models are correct, 190,000 people in Wuhan will be infected by now, major Chinese cities are on the cusp of large-scale outbreaks, and the virus will continue to spread to other countries.

Finding the Start
The spread of a given virus is partly linked to how long it remains undetected. Identifying a new virus is the first step towards mobilizing a response and, in time, creating a vaccine. Warning at-risk populations as quickly as possible also helps with limiting the spread.

These are among the reasons why BlueDot’s achievement is important in and of itself. Furthermore, it illustrates how AIs can sift through vast troves of data to identify ongoing virus outbreaks.

BlueDot uses natural language processing and machine learning to scour a variety of information sources, including chomping through 100,000 news reports in 65 languages a day. Data is compared with flight records to help predict virus outbreak patterns. Once the automated data sifting is completed, epidemiologists check that the findings make sense from a scientific standpoint, and reports are sent to BlueDot’s customers, which include governments, businesses, and public health organizations.

AI for Virus Detection and Prevention
Other companies, such as Metabiota, are also using data-driven approaches to track the spread of the likes of the coronavirus.

Researchers have trained neural networks to predict the spread of infectious diseases in real time. Others are using AI algorithms to identify how preventive measures can have the greatest effect. AI is also being used to create new drugs, which we may well see repeated for the coronavirus.

If the work of scientists Barbara Han and David Redding comes to fruition, AI and machine learning may even help us predict where virus outbreaks are likely to strike—before they do.

The Uncertainty Factor
One of AI’s core strengths when working on identifying and limiting the effects of virus outbreaks is its incredibly insistent nature. AIs never tire, can sift through enormous amounts of data, and identify possible correlations and causations that humans can’t.

However, there are limits to AI’s ability to both identify virus outbreaks and predict how they will spread. Perhaps the best-known example comes from the neighboring field of big data analytics. At its launch, Google Flu Trends was heralded as a great leap forward in relation to identifying and estimating the spread of the flu—until it underestimated the 2013 flu season by a whopping 140 percent and was quietly put to rest.

Poor data quality was identified as one of the main reasons Google Flu Trends failed. Unreliable or faulty data can wreak havoc on the prediction power of AIs.

In our increasingly interconnected world, tracking the movements of potentially infected individuals (by car, trains, buses, or planes) is just one vector surrounded by a lot of uncertainty.

The fact that BlueDot was able to correctly identify the coronavirus, in part due to its AI technology, illustrates that smart computer systems can be incredibly useful in helping us navigate these uncertainties.

Importantly, though, this isn’t the same as AI being at a point where it unerringly does so on its own—which is why BlueDot employs human experts to validate the AI’s findings.

Image Credit: Coronavirus molecular illustration, Gianluca Tomasello/Wikimedia Commons Continue reading

Posted in Human Robots

#435492 Humanoid table tennis players

Trust the Chinese to come up with Android Robots for one of their favorite sports! The robot can also play a human opponent, using either forehand or backhand strokes. Related Posts 9 Robot Animals Built From Nature’s …Millions of years … Continue reading

Posted in Human Robots

#435806 Boston Dynamics’ Spot Robot Dog ...

Boston Dynamics is announcing this morning that Spot, its versatile quadruped robot, is now for sale. The machine’s animal-like behavior regularly electrifies crowds at tech conferences, and like other Boston Dynamics’ robots, Spot is a YouTube sensation whose videos amass millions of views.

Now anyone interested in buying a Spot—or a pack of them—can go to the company’s website and submit an order form. But don’t pull out your credit card just yet. Spot may cost as much as a luxury car, and it is not really available to consumers. The initial sale, described as an “early adopter program,” is targeting businesses. Boston Dynamics wants to find customers in select industries and help them deploy Spots in real-world scenarios.

“What we’re doing is the productization of Spot,” Boston Dynamics CEO Marc Raibert tells IEEE Spectrum. “It’s really a milestone for us going from robots that work in the lab to these that are hardened for work out in the field.”

Boston Dynamics has always been a secretive company, but last month, in preparation for launching Spot (formerly SpotMini), it allowed our photographers into its headquarters in Waltham, Mass., for a special shoot. In that session, we captured Spot and also Atlas—the company’s highly dynamic humanoid—in action, walking, climbing, and jumping.

You can see Spot’s photo interactives on our Robots Guide. (The Atlas interactives will appear in coming weeks.)

Gif: Bob O’Connor/Robots.ieee.org

And if you’re in the market for a robot dog, here’s everything we know about Boston Dynamics’ plans for Spot.

Who can buy a Spot?
If you’re interested in one, you should go to Boston Dynamics’ website and take a look at the information the company requires from potential buyers. Again, the focus is on businesses. Boston Dynamics says it wants to get Spots out to initial customers that “either have a compelling use case or a development team that we believe can do something really interesting with the robot,” says VP of business development Michael Perry. “Just because of the scarcity of the robots that we have, we’re going to have to be selective about which partners we start working together with.”

What can Spot do?
As you’ve probably seen on the YouTube videos, Spot can walk, trot, avoid obstacles, climb stairs, and much more. The robot’s hardware is almost completely custom, with powerful compute boards for control, and five sensor modules located on every side of Spot’s body, allowing it to survey the space around itself from any direction. The legs are powered by 12 custom motors with a reduction, with a top speed of 1.6 meters per second. The robot can operate for 90 minutes on a charge. In addition to the basic configuration, you can integrate up to 14 kilograms of extra hardware to a payload interface. Among the payload packages Boston Dynamics plans to offer are a 6 degrees-of-freedom arm, a version of which can be seen in some of the YouTube videos, and a ring of cameras called SpotCam that could be used to create Street View–type images inside buildings.

Image: Boston Dynamics

How do you control Spot?
Learning to drive the robot using its gaming-style controller “takes 15 seconds,” says CEO Marc Raibert. He explains that while teleoperating Spot, you may not realize that the robot is doing a lot of the work. “You don’t really see what that is like until you’re operating the joystick and you go over a box and you don’t have to do anything,” he says. “You’re practically just thinking about what you want to do and the robot takes care of everything.” The control methods have evolved significantly since the company’s first quadruped robots, machines like BigDog and LS3. “The control in those days was much more monolithic, and now we have what we call a sequential composition controller,” Raibert says, “which lets the system have control of the dynamics in a much broader variety of situations.” That means that every time one of Spot’s feet touches or doesn’t touch the ground, this different state of the body affects the basic physical behavior of the robot, and the controller adjusts accordingly. “Our controller is designed to understand what that state is and have different controls depending upon the case,” he says.

How much does Spot cost?
Boston Dynamics would not give us specific details about pricing, saying only that potential customers should contact them for a quote and that there is going to be a leasing option. It’s understandable: As with any expensive and complex product, prices can vary on a case by case basis and depend on factors such as configuration, availability, level of support, and so forth. When we pressed the company for at least an approximate base price, Perry answered: “Our general guidance is that the total cost of the early adopter program lease will be less than the price of a car—but how nice a car will depend on the number of Spots leased and how long the customer will be leasing the robot.”

Can Spot do mapping and SLAM out of the box?
The robot’s perception system includes cameras and 3D sensors (there is no lidar), used to avoid obstacles and sense the terrain so it can climb stairs and walk over rubble. It’s also used to create 3D maps. According to Boston Dynamics, the first software release will offer just teleoperation. But a second release, to be available in the next few weeks, will enable more autonomous behaviors. For example, it will be able to do mapping and autonomous navigation—similar to what the company demonstrated in a video last year, showing how you can drive the robot through an environment, create a 3D point cloud of the environment, and then set waypoints within that map for Spot to go out and execute that mission. For customers that have their own autonomy stack and are interested in using those on Spot, Boston Dynamics made it “as plug and play as possible in terms of how third-party software integrates into Spot’s system,” Perry says. This is done mainly via an API.

How does Spot’s API works?
Boston Dynamics built an API so that customers can create application-level products with Spot without having to deal with low-level control processes. “Rather than going and building joint-level kinematic access to the robot,” Perry explains, “we created a high-level API and SDK that allows people who are used to Web app development or development of missions for drones to use that same scope, and they’ll be able to build applications for Spot.”

What applications should we see first?
Boston Dynamics envisions Spot as a platform: a versatile mobile robot that companies can use to build applications based on their needs. What types of applications? The company says the best way to find out is to put Spot in the hands of as many users as possible and let them develop the applications. Some possibilities include performing remote data collection and light manipulation in construction sites; monitoring sensors and infrastructure at oil and gas sites; and carrying out dangerous missions such as bomb disposal and hazmat inspections. There are also other promising areas such as security, package delivery, and even entertainment. “We have some initial guesses about which markets could benefit most from this technology, and we’ve been engaging with customers doing proof-of-concept trials,” Perry says. “But at the end of the day, that value story is really going to be determined by people going out and exploring and pushing the limits of the robot.”

Photo: Bob O'Connor

How many Spots have been produced?
Last June, Boston Dynamics said it was planning to build about a hundred Spots by the end of the year, eventually ramping up production to a thousand units per year by the middle of this year. The company admits that it is not quite there yet. It has built close to a hundred beta units, which it has used to test and refine the final design. This version is now being mass manufactured, but the company is still “in the early tens of robots,” Perry says.

How did Boston Dynamics test Spot?

The company has tested the robots during proof-of-concept trials with customers, and at least one is already using Spot to survey construction sites. The company has also done reliability tests at its facility in Waltham, Mass. “We drive around, not quite day and night, but hundreds of miles a week, so that we can collect reliability data and find bugs,” Raibert says.

What about competitors?
In recent years, there’s been a proliferation of quadruped robots that will compete in the same space as Spot. The most prominent of these is ANYmal, from ANYbotics, a Swiss company that spun out of ETH Zurich. Other quadrupeds include Vision from Ghost Robotics, used by one of the teams in the DARPA Subterranean Challenge; and Laikago and Aliengo from Unitree Robotics, a Chinese startup. Raibert views the competition as a positive thing. “We’re excited to see all these companies out there helping validate the space,” he says. “I think we’re more in competition with finding the right need [that robots can satisfy] than we are with the other people building the robots at this point.”

Why is Boston Dynamics selling Spot now?
Boston Dynamics has long been an R&D-centric firm, with most of its early funding coming from military programs, but it says commercializing robots has always been a goal. Productizing its machines probably accelerated when the company was acquired by Google’s parent company, Alphabet, which had an ambitious (and now apparently very dead) robotics program. The commercial focus likely continued after Alphabet sold Boston Dynamics to SoftBank, whose famed CEO, Masayoshi Son, is known for his love of robots—and profits.

Which should I buy, Spot or Aibo?
Don’t laugh. We’ve gotten emails from individuals interested in purchasing a Spot for personal use after seeing our stories on the robot. Alas, Spot is not a bigger, fancier Aibo pet robot. It’s an expensive, industrial-grade machine that requires development and maintenance. If you’re maybe Jeff Bezos you could probably convince Boston Dynamics to sell you one, but otherwise the company will prioritize businesses.

What’s next for Boston Dynamics?
On the commercial side of things, other than Spot, Boston Dynamics is interested in the logistics space. Earlier this year it announced the acquisition of Kinema Systems, a startup that had developed vision sensors and deep-learning software to enable industrial robot arms to locate and move boxes. There’s also Handle, the mobile robot on whegs (wheels + legs), that can pick up and move packages. Boston Dynamics is hiring both in Waltham, Mass., and Mountain View, Calif., where Kinema was located.

Okay, can I watch a cool video now?
During our visit to Boston Dynamics’ headquarters last month, we saw Atlas and Spot performing some cool new tricks that we unfortunately are not allowed to tell you about. We hope that, although the company is putting a lot of energy and resources into its commercial programs, Boston Dynamics will still find plenty of time to improve its robots, build new ones, and of course, keep making videos. [Update: The company has just released a new Spot video, which we’ve embedded at the top of the post.][Update 2: We should have known. Boston Dynamics sure knows how to create buzz for itself: It has just released a second video, this time of Atlas doing some of those tricks we saw during our visit and couldn’t tell you about. Enjoy!]

[ Boston Dynamics ] Continue reading

Posted in Human Robots

#435474 Watch China’s New Hybrid AI Chip Power ...

When I lived in Beijing back in the 90s, a man walking his bike was nothing to look at. But today, I did a serious double-take at a video of a bike walking his man.

No kidding.

The bike itself looks overloaded but otherwise completely normal. Underneath its simplicity, however, is a hybrid computer chip that combines brain-inspired circuits with machine learning processes into a computing behemoth. Thanks to its smart chip, the bike self-balances as it gingerly rolls down a paved track before smoothly gaining speed into a jogging pace while navigating dexterously around obstacles. It can even respond to simple voice commands such as “speed up,” “left,” or “straight.”

Far from a circus trick, the bike is a real-world demo of the AI community’s latest attempt at fashioning specialized hardware to keep up with the challenges of machine learning algorithms. The Tianjic (天机*) chip isn’t just your standard neuromorphic chip. Rather, it has the architecture of a brain-like chip, but can also run deep learning algorithms—a match made in heaven that basically mashes together neuro-inspired hardware and software.

The study shows that China is readily nipping at the heels of Google, Facebook, NVIDIA, and other tech behemoths investing in developing new AI chip designs—hell, with billions in government investment it may have already had a head start. A sweeping AI plan from 2017 looks to catch up with the US on AI technology and application by 2020. By 2030, China’s aiming to be the global leader—and a champion for building general AI that matches humans in intellectual competence.

The country’s ambition is reflected in the team’s parting words.

“Our study is expected to stimulate AGI [artificial general intelligence] development by paving the way to more generalized hardware platforms,” said the authors, led by Dr. Luping Shi at Tsinghua University.

A Hardware Conundrum
Shi’s autonomous bike isn’t the first robotic two-wheeler. Back in 2015, the famed research nonprofit SRI International in Menlo Park, California teamed up with Yamaha to engineer MOTOBOT, a humanoid robot capable of driving a motorcycle. Powered by state-of-the-art robotic hardware and machine learning, MOTOBOT eventually raced MotoGPTM world champion Valentino Rossi in a nail-biting match-off.

However, the technological core of MOTOBOT and Shi’s bike vastly differ, and that difference reflects two pathways towards more powerful AI. One, exemplified by MOTOBOT, is software—developing brain-like algorithms with increasingly efficient architecture, efficacy, and speed. That sounds great, but deep neural nets demand so many computational resources that general-purpose chips can’t keep up.

As Shi told China Science Daily: “CPUs and other chips are driven by miniaturization technologies based on physics. Transistors might shrink to nanoscale-level in 10, 20 years. But what then?” As more transistors are squeezed onto these chips, efficient cooling becomes a limiting factor in computational speed. Tax them too much, and they melt.

For AI processes to continue, we need better hardware. An increasingly popular idea is to build neuromorphic chips, which resemble the brain from the ground up. IBM’s TrueNorth, for example, contains a massively parallel architecture nothing like the traditional Von Neumann structure of classic CPUs and GPUs. Similar to biological brains, TrueNorth’s memory is stored within “synapses” between physical “neurons” etched onto the chip, which dramatically cuts down on energy consumption.

But even these chips are limited. Because computation is tethered to hardware architecture, most chips resemble just one specific type of brain-inspired network called spiking neural networks (SNNs). Without doubt, neuromorphic chips are highly efficient setups with dynamics similar to biological networks. They also don’t play nicely with deep learning and other software-based AI.

Brain-AI Hybrid Core
Shi’s new Tianjic chip brought the two incompatibilities together onto a single piece of brainy hardware.

First was to bridge the deep learning and SNN divide. The two have very different computation philosophies and memory organizations, the team said. The biggest difference, however, is that artificial neural networks transform multidimensional data—image pixels, for example—into a single, continuous, multi-bit 0 and 1 stream. In contrast, neurons in SNNs activate using something called “binary spikes” that code for specific activation events in time.

Confused? Yeah, it’s hard to wrap my head around it too. That’s because SNNs act very similarly to our neural networks and nothing like computers. A particular neuron needs to generate an electrical signal (a “spike”) large enough to transfer down to the next one; little blips in signals don’t count. The way they transmit data also heavily depends on how they’re connected, or the network topology. The takeaway: SNNs work pretty differently than deep learning.

Shi’s team first recreated this firing quirk in the language of computers—0s and 1s—so that the coding mechanism would become compatible with deep learning algorithms. They then carefully aligned the step-by-step building blocks of the two models, which allowed them to tease out similarities into a common ground to further build on. “On the basis of this unified abstraction, we built a cross-paradigm neuron scheme,” they said.

In general, the design allowed both computational approaches to share the synapses, where neurons connect and store data, and the dendrites, the outgoing branches of the neurons. In contrast, the neuron body, where signals integrate, was left reconfigurable for each type of computation, as were the input branches. Each building block was combined into a single unified functional core (FCore), which acts like a deep learning/SNN converter depending on its specific setup. Translation: the chip can do both types of previously incompatible computation.

The Chip
Using nanoscale fabrication, the team arranged 156 FCores, containing roughly 40,000 neurons and 10 million synapses, onto a chip less than a fifth of an inch in length and width. Initial tests showcased the chip’s versatility, in that it can run both SNNs and deep learning algorithms such as the popular convolutional neural network (CNNs) often used in machine vision.

Compared to IBM TrueNorth, the density of Tianjic’s cores increased by 20 percent, speeding up performance ten times and increasing bandwidth at least 100-fold, the team said. When pitted against GPUs, the current hardware darling of machine learning, the chip increased processing throughput up to 100 times, while using just a sliver (1/10,000) of energy.

Although these stats are great, real-life performance is even better as a demo. Here’s where the authors gave their Tianjic brain a body. The team combined one chip with multiple specialized networks to process vision, balance, voice commands, and decision-making in real time. Object detection and target tracking, for example, relied on a deep neural net CNN, whereas voice commands and balance data were recognized using an SNN. The inputs were then integrated inside a neural state machine, which churned out decisions to downstream output modules—for example, controlling the handle bar to turn left.

Thanks to the chip’s brain-like architecture and bilingual ability, Tianjic “allowed all of the neural network models to operate in parallel and realized seamless communication across the models,” the team said. The result is an autonomous bike that rolls after its human, balances across speed bumps, avoids crashing into roadblocks, and answers to voice commands.

General AI?
“It’s a wonderful demonstration and quite impressive,” said the editorial team at Nature, which published the study on its cover last week.

However, they cautioned, when comparing Tianjic with state-of-the-art chips designed for a single problem toe-to-toe on that particular problem, Tianjic falls behind. But building these jack-of-all-trades hybrid chips is definitely worth the effort. Compared to today’s limited AI, what people really want is artificial general intelligence, which will require new architectures that aren’t designed to solve one particular problem.

Until people start to explore, innovate, and play around with different designs, it’s not clear how we can further progress in the pursuit of general AI. A self-driving bike might not be much to look at, but its hybrid brain is a pretty neat place to start.

*The name, in Chinese, means “heavenly machine,” “unknowable mystery of nature,” or “confidentiality.” Go figure.

Image Credit: Alexander Ryabintsev / Shutterstock.com Continue reading

Posted in Human Robots