Tag Archives: applied
Con artistry is one of the world’s oldest and most innovative professions, and it may soon have a new target. Research suggests artificial intelligence may be uniquely susceptible to tricksters, and as its influence in the modern world grows, attacks against it are likely to become more common.
The root of the problem lies in the fact that artificial intelligence algorithms learn about the world in very different ways than people do, and so slight tweaks to the data fed into these algorithms can throw them off completely while remaining imperceptible to humans.
Much of the research into this area has been conducted on image recognition systems, in particular those relying on deep learning neural networks. These systems are trained by showing them thousands of examples of images of a particular object until they can extract common features that allow them to accurately spot the object in new images.
But the features they extract are not necessarily the same high-level features a human would be looking for, like the word STOP on a sign or a tail on a dog. These systems analyze images at the individual pixel level to detect patterns shared between examples. These patterns can be obscure combinations of pixel values, in small pockets or spread across the image, that would be impossible to discern for a human, but highly accurate at predicting a particular object.
“An attacker can trick the object recognition algorithm into seeing something that isn’t there, without these alterations being obvious to a human.”
What this means is that by identifying these patterns and overlaying them over a different image, an attacker can trick the object recognition algorithm into seeing something that isn’t there, without these alterations being obvious to a human. This kind of manipulation is known as an “adversarial attack.”
Early attempts to trick image recognition systems this way required access to the algorithm’s inner workings to decipher these patterns. But in 2016 researchers demonstrated a “black box” attack that enabled them to trick such a system without knowing its inner workings.
By feeding the system doctored images and seeing how it classified them, they were able to work out what it was focusing on and therefore generate images they knew would fool it. Importantly, the doctored images were not obviously different to human eyes.
These approaches were tested by feeding doctored image data directly into the algorithm, but more recently, similar approaches have been applied in the real world. Last year it was shown that printouts of doctored images that were then photographed on a smartphone successfully tricked an image classification system.
Another group showed that wearing specially designed, psychedelically-colored spectacles could trick a facial recognition system into thinking people were celebrities. In August scientists showed that adding stickers to stop signs in particular configurations could cause a neural net designed to spot them to misclassify the signs.
These last two examples highlight some of the potential nefarious applications for this technology. Getting a self-driving car to miss a stop sign could cause an accident, either for insurance fraud or to do someone harm. If facial recognition becomes increasingly popular for biometric security applications, being able to pose as someone else could be very useful to a con artist.
Unsurprisingly, there are already efforts to counteract the threat of adversarial attacks. In particular, it has been shown that deep neural networks can be trained to detect adversarial images. One study from the Bosch Center for AI demonstrated such a detector, an adversarial attack that fools the detector, and a training regime for the detector that nullifies the attack, hinting at the kind of arms race we are likely to see in the future.
While image recognition systems provide an easy-to-visualize demonstration, they’re not the only machine learning systems at risk. The techniques used to perturb pixel data can be applied to other kinds of data too.
“Bypassing cybersecurity defenses is one of the more worrying and probable near-term applications for this approach.”
Chinese researchers showed that adding specific words to a sentence or misspelling a word can completely throw off machine learning systems designed to analyze what a passage of text is about. Another group demonstrated that garbled sounds played over speakers could make a smartphone running the Google Now voice command system visit a particular web address, which could be used to download malware.
This last example points toward one of the more worrying and probable near-term applications for this approach: bypassing cybersecurity defenses. The industry is increasingly using machine learning and data analytics to identify malware and detect intrusions, but these systems are also highly susceptible to trickery.
At this summer’s DEF CON hacking convention, a security firm demonstrated they could bypass anti-malware AI using a similar approach to the earlier black box attack on the image classifier, but super-powered with an AI of their own.
Their system fed malicious code to the antivirus software and then noted the score it was given. It then used genetic algorithms to iteratively tweak the code until it was able to bypass the defenses while maintaining its function.
All the approaches noted so far are focused on tricking pre-trained machine learning systems, but another approach of major concern to the cybersecurity industry is that of “data poisoning.” This is the idea that introducing false data into a machine learning system’s training set will cause it to start misclassifying things.
This could be particularly challenging for things like anti-malware systems that are constantly being updated to take into account new viruses. A related approach bombards systems with data designed to generate false positives so the defenders recalibrate their systems in a way that then allows the attackers to sneak in.
How likely it is that these approaches will be used in the wild will depend on the potential reward and the sophistication of the attackers. Most of the techniques described above require high levels of domain expertise, but it’s becoming ever easier to access training materials and tools for machine learning.
Simpler versions of machine learning have been at the heart of email spam filters for years, and spammers have developed a host of innovative workarounds to circumvent them. As machine learning and AI increasingly embed themselves in our lives, the rewards for learning how to trick them will likely outweigh the costs.
Image Credit: Nejron Photo / Shutterstock.com Continue reading
It is one of the top 10 deadliest diseases in the United States, and it cannot be cured or prevented. But new studies are finding ways to diagnose Alzheimer’s disease in its earliest stages, while some of the latest research says technologies like artificial intelligence can detect dementia years before the first symptoms occur.
These advances, in turn, will help bolster clinical trials seeking a cure or therapies to slow or prevent the disease. Catching Alzheimer’s disease or other forms of dementia early in their progression can help ease symptoms in some cases.
“Often neurodegeneration is diagnosed late when massive brain damage has already occurred,” says professor Francis L Martin at the University of Central Lancashire in the UK, in an email to Singularity Hub. “As we know more about the molecular basis of the disease, there is the possibility of clinical interventions that might slow or halt the progress of the disease, i.e., before brain damage. Extending cognitive ability for even a number of years would have huge benefit.”
Martin is the principal investigator on a project that has developed a technique to analyze blood samples to diagnose Alzheimer’s disease and distinguish between other forms of dementia.
The researchers used sensor-based technology with a diamond core to analyze about 550 blood samples. They identified specific chemical bonds within the blood after passing light through the diamond core and recording its interaction with the sample. The results were then compared against blood samples from cases of Alzheimer’s disease and other neurodegenerative diseases, along with those from healthy individuals.
“From a small drop of blood, we derive a fingerprint spectrum. That fingerprint spectrum contains numerical data, which can be inputted into a computational algorithm we have developed,” Martin explains. “This algorithm is validated for prediction of unknown samples. From this we determine sensitivity and specificity. Although not perfect, my clinical colleagues reliably tell me our results are far better than anything else they have seen.”
Martin says the breakthrough is the result of more than 10 years developing sensor-based technologies for routine screening, monitoring, or diagnosing neurodegenerative diseases and cancers.
“My vision was to develop something low-cost that could be readily applied in a typical clinical setting to handle thousands of samples potentially per day or per week,” he says, adding that the technology also has applications in environmental science and food security.
The new test can also distinguish accurately between Alzheimer’s disease and other forms of neurodegeneration, such as Lewy body dementia, which is one of the most common causes of dementia after Alzheimer’s.
“To this point, other than at post-mortem, there has been no single approach towards classifying these pathologies,” Martin notes. “MRI scanning is often used but is labor-intensive, costly, difficult to apply to dementia patients, and not a routine point-of-care test.”
Canadian researchers at McGill University believe they can predict Alzheimer’s disease up to two years before its onset using big data and artificial intelligence. They developed an algorithm capable of recognizing the signatures of dementia using a single amyloid PET scan of the brain of patients at risk of developing the disease.
Alzheimer’s is caused by the accumulation of two proteins—amyloid beta and tau. The latest research suggests that amyloid beta leads to the buildup of tau, which is responsible for damaging nerve cells and connections between cells called synapses.
The work was recently published in the journal Neurobiology of Aging.
“Despite the availability of biomarkers capable of identifying the proteins causative of Alzheimer’s disease in living individuals, the current technologies cannot predict whether carriers of AD pathology in the brain will progress to dementia,” Sulantha Mathotaarachchi, lead author on the paper and an expert in artificial neural networks, tells Singularity Hub by email.
The algorithm, trained on a population with amnestic mild cognitive impairment observed over 24 months, proved accurate 84.5 percent of the time. Mathotaarachchi says the algorithm can be trained on different populations for different observational periods, meaning the system can grow more comprehensive with more data.
“The more biomarkers we incorporate, the more accurate the prediction could be,” Mathotaarachchi adds. “However, right now, acquiring [the] required amount of training data is the biggest challenge. … In Alzheimer’s disease, it is known that the amyloid protein deposition occurs decades before symptoms onset.”
Unfortunately, the same process occurs in normal aging as well. “The challenge is to identify the abnormal patterns of deposition that lead to the disease later on,” he says
One of the key goals of the project is to improve the research in Alzheimer’s disease by ensuring those patients with the highest probability to develop dementia are enrolled in clinical trials. That will increase the efficiency of clinical programs, according to Mathotaarachchi.
“One of the most important outcomes from our study was the pilot, online, real-time prediction tool,” he says. “This can be used as a framework for patient screening before recruiting for clinical trials. … If a disease-modifying therapy becomes available for patients, a predictive tool might have clinical applications as well, by providing to the physician information regarding clinical progression.”
Pixel by Pixel Prediction
Private industry is also working toward improving science’s predictive powers when it comes to detecting dementia early. One startup called Darmiyan out of San Francisco claims its proprietary software can pick up signals before the onset of Alzheimer’s disease by up to 15 years.
Darmiyan didn’t respond to a request for comment for this article. Venture Beat reported that the company’s MRI-analyzing software “detects cell abnormalities at a microscopic level to reveal what a standard MRI scan cannot” and that the “software measures and highlights subtle microscopic changes in the brain tissue represented in every pixel of the MRI image long before any symptoms arise.”
Darmiyan claims to have a 90 percent accuracy rate and says its software has been vetted by top academic institutions like New York University, Rockefeller University, and Stanford, according to Venture Beat. The startup is awaiting FDA approval to proceed further but is reportedly working with pharmaceutical companies like Amgen, Johnson & Johnson, and Pfizer on pilot programs.
“Our technology enables smarter drug selection in preclinical animal studies, better patient selection for clinical trials, and much better drug-effect monitoring,” Darmiyan cofounder and CEO Padideh Kamali-Zare told Venture Beat.
An estimated 5.5 million Americans have Alzheimer’s, and one in 10 people over age 65 have been diagnosed with the disease. By mid-century, the number of Alzheimer’s patients could rise to 16 million. Health care costs in 2017 alone are estimated to be $259 billion, and by 2050 the annual price tag could be more than $1 trillion.
In sum, it’s a disease that cripples people and the economy.
Researchers are always after more data as they look to improve outcomes, with the hope of one day developing a cure or preventing the onset of neurodegeneration altogether. If interested in seeing this medical research progress, you can help by signing up on the Brain Health Registry to improve the quality of clinical trials.
Image Credit: rudall30 / Shutterstock.com Continue reading
The brain has long inspired the design of computers and their software. Now Intel has become the latest tech company to decide that mimicking the brain’s hardware could be the next stage in the evolution of computing.
On Monday the company unveiled an experimental “neuromorphic” chip called Loihi. Neuromorphic chips are microprocessors whose architecture is configured to mimic the biological brain’s network of neurons and the connections between them called synapses.
While neural networks—the in vogue approach to artificial intelligence and machine learning—are also inspired by the brain and use layers of virtual neurons, they are still implemented on conventional silicon hardware such as CPUs and GPUs.
The main benefit of mimicking the architecture of the brain on a physical chip, say neuromorphic computing’s proponents, is energy efficiency—the human brain runs on roughly 20 watts. The “neurons” in neuromorphic chips carry out the role of both processor and memory which removes the need to shuttle data back and forth between separate units, which is how traditional chips work. Each neuron also only needs to be powered while it’s firing.
At present, most machine learning is done in data centers due to the massive energy and computing requirements. Creating chips that capture some of nature’s efficiency could allow AI to be run directly on devices like smartphones, cars, and robots.
This is exactly the kind of application Michael Mayberry, managing director of Intel’s research arm, touts in a blog post announcing Loihi. He talks about CCTV cameras that can run image recognition to identify missing persons or traffic lights that can track traffic flow to optimize timing and keep vehicles moving.
There’s still a long way to go before that happens though. According to Wired, so far Intel has only been working with prototypes, and the first full-size version of the chip won’t be built until November.
Once complete, it will feature 130,000 neurons and 130 million synaptic connections split between 128 computing cores. The device will be 1,000 times more energy-efficient than standard approaches, according to Mayberry, but more impressive are claims the chip will be capable of continuous learning.
Intel’s newly launched self-learning neuromorphic chip.
Normally deep learning works by training a neural network on giant datasets to create a model that can then be applied to new data. The Loihi chip will combine training and inference on the same chip, which will allow it to learn on the fly, constantly updating its models and adapting to changing circumstances without having to be deliberately re-trained.
A select group of universities and research institutions will be the first to get their hands on the new chip in the first half of 2018, but Mayberry said it could be years before it’s commercially available. Whether commercialization happens at all may largely depend on whether early adopters can get the hardware to solve any practically useful problems.
So far neuromorphic computing has struggled to gain traction outside the research community. IBM released a neuromorphic chip called TrueNorth in 2014, but the device has yet to showcase any commercially useful applications.
Lee Gomes summarizes the hurdles facing neuromorphic computing excellently in IEEE Spectrum. One is that deep learning can run on very simple, low-precision hardware that can be optimized to use very little power, which suggests complicated new architectures may struggle to find purchase.
It’s also not easy to transfer deep learning approaches developed on conventional chips over to neuromorphic hardware, and even Intel Labs chief scientist Narayan Srinivasa admitted to Forbes Loihi wouldn’t work well with some deep learning models.
Finally, there’s considerable competition in the quest to develop new computer architectures specialized for machine learning. GPU vendors Nvidia and AMD have pivoted to take advantage of this newfound market and companies like Google and Microsoft are developing their own in-house solutions.
Intel, for its part, isn’t putting all its eggs in one basket. Last year it bought two companies building chips for specialized machine learning—Movidius and Nervana—and this was followed up with the $15 billion purchase of self-driving car chip- and camera-maker Mobileye.
And while the jury is still out on neuromorphic computing, it makes sense for a company eager to position itself as the AI chipmaker of the future to have its fingers in as many pies as possible. There are a growing number of voices suggesting that despite its undoubted power, deep learning alone will not allow us to imbue machines with the kind of adaptable, general intelligence humans possess.
What new approaches will get us there are hard to predict, but it’s entirely possible they will only work on hardware that closely mimics the one device we already know is capable of supporting this kind of intelligence—the human brain.
Image Credit: Intel Continue reading
As Dorothy famously said in The Wizard of Oz, there’s no place like home. Home is where we go to rest and recharge. It’s familiar, comfortable, and our own. We take care of our homes by cleaning and maintaining them, and fixing things that break or go wrong.
What if our homes, on top of giving us shelter, could also take care of us in return?
According to Chris Arkenberg, this could be the case in the not-so-distant future. As part of Singularity University’s Experts On Air series, Arkenberg gave a talk called “How the Intelligent Home of The Future Will Care For You.”
Arkenberg is a research and strategy lead at Orange Silicon Valley, and was previously a research fellow at the Deloitte Center for the Edge and a visiting researcher at the Institute for the Future.
Arkenberg told the audience that there’s an evolution going on: homes are going from being smart to being connected, and will ultimately become intelligent.
Intelligent home technologies are just now budding, but broader trends point to huge potential for their growth. We as consumers already expect continuous connectivity wherever we go—what do you mean my phone won’t get reception in the middle of Yosemite? What do you mean the smart TV is down and I can’t stream Game of Thrones?
As connectivity has evolved from a privilege to a basic expectation, Arkenberg said, we’re also starting to have a better sense of what it means to give up our data in exchange for services and conveniences. It’s so easy to click a few buttons on Amazon and have stuff show up at your front door a few days later—never mind that data about your purchases gets recorded and aggregated.
“Right now we have single devices that are connected,” Arkenberg said. “Companies are still trying to show what the true value is and how durable it is beyond the hype.”
Connectivity is the basis of an intelligent home. To take a dumb object and make it smart, you get it online. Belkin’s Wemo, for example, lets users control lights and appliances wirelessly and remotely, and can be paired with Amazon Echo or Google Home for voice-activated control.
Speaking of voice-activated control, Arkenberg pointed out that physical interfaces are evolving, too, to the point that we’re actually getting rid of interfaces entirely, or transitioning to ‘soft’ interfaces like voice or gesture.
Drivers of change
Consumers are open to smart home tech and companies are working to provide it. But what are the drivers making this tech practical and affordable? Arkenberg said there are three big ones:
Computation: Computers have gotten exponentially more powerful over the past few decades. If it wasn’t for processors that could handle massive quantities of information, nothing resembling an Echo or Alexa would even be possible. Artificial intelligence and machine learning are powering these devices, and they hinge on computing power too.
Sensors: “There are more things connected now than there are people on the planet,” Arkenberg said. Market research firm Gartner estimates there are 8.4 billion connected things currently in use. Wherever digital can replace hardware, it’s doing so. Cheaper sensors mean we can connect more things, which can then connect to each other.
Data: “Data is the new oil,” Arkenberg said. “The top companies on the planet are all data-driven giants. If data is your business, though, then you need to keep finding new ways to get more and more data.” Home assistants are essentially data collection systems that sit in your living room and collect data about your life. That data in turn sets up the potential of machine learning.
Colonizing the Living Room
Alexa and Echo can turn lights on and off, and Nest can help you be energy-efficient. But beyond these, what does an intelligent home really look like?
Arkenberg’s vision of an intelligent home uses sensing, data, connectivity, and modeling to manage resource efficiency, security, productivity, and wellness.
Autonomous vehicles provide an interesting comparison: they’re surrounded by sensors that are constantly mapping the world to build dynamic models to understand the change around itself, and thereby predict things. Might we want this to become a model for our homes, too? By making them smart and connecting them, Arkenberg said, they’d become “more biological.”
There are already several products on the market that fit this description. RainMachine uses weather forecasts to adjust home landscape watering schedules. Neurio monitors energy usage, identifies areas where waste is happening, and makes recommendations for improvement.
These are small steps in connecting our homes with knowledge systems and giving them the ability to understand and act on that knowledge.
He sees the homes of the future being equipped with digital ears (in the form of home assistants, sensors, and monitoring devices) and digital eyes (in the form of facial recognition technology and machine vision to recognize who’s in the home). “These systems are increasingly able to interrogate emotions and understand how people are feeling,” he said. “When you push more of this active intelligence into things, the need for us to directly interface with them becomes less relevant.”
Could our homes use these same tools to benefit our health and wellness? FREDsense uses bacteria to create electrochemical sensors that can be applied to home water systems to detect contaminants. If that’s not personal enough for you, get a load of this: ClinicAI can be installed in your toilet bowl to monitor and evaluate your biowaste. What’s the point, you ask? Early detection of colon cancer and other diseases.
What if one day, your toilet’s biowaste analysis system could link up with your fridge, so that when you opened it it would tell you what to eat, and how much, and at what time of day?
Roadblocks to intelligence
“The connected and intelligent home is still a young category trying to establish value, but the technological requirements are now in place,” Arkenberg said. We’re already used to living in a world of ubiquitous computation and connectivity, and we have entrained expectations about things being connected. For the intelligent home to become a widespread reality, its value needs to be established and its challenges overcome.
One of the biggest challenges will be getting used to the idea of continuous surveillance. We’ll get convenience and functionality if we give up our data, but how far are we willing to go? Establishing security and trust is going to be a big challenge moving forward,” Arkenberg said.
There’s also cost and reliability, interoperability and fragmentation of devices, or conversely, what Arkenberg called ‘platform lock-on,’ where you’d end up relying on only one provider’s system and be unable to integrate devices from other brands.
Ultimately, Arkenberg sees homes being able to learn about us, manage our scheduling and transit, watch our moods and our preferences, and optimize our resource footprint while predicting and anticipating change.
“This is the really fascinating provocation of the intelligent home,” Arkenberg said. “And I think we’re going to start to see this play out over the next few years.”
Sounds like a home Dorothy wouldn’t recognize, in Kansas or anywhere else.
Stock Media provided by adam121 / Pond5 Continue reading