Slashdot Mirror

RockDoctor writes: An ad hoc committee of the International Astronomical Union has been working for 5 months on revisions and clarifications to the definitions of various terms used in technical and popular discussions of SETI -- the Search For Extraterrestrial Intelligence. They've published their draft report.

The terms of reference -- to account for existing popular and technical uses of the terms -- should mean that no major changes in usage occur, but interesting points do emerge from the discussion paper. For example, in discussing the term "extraterrestrial," their proposed definition ("shorthand for life or technology not originating recently on Earth") includes cover for possibilities such as "panspermia" which may be popular in "popular science," but certainly are not popular in the technical discussions. They go on to discuss that "by this definition, life on another planet with a common origin to Earth life but which diverged billions of years ago would be extraterrestrial, but Earth life accidentally brought to Mars on a human-built lander would not." Waiting for the invasion of the pedants, clutching their feet in their hands.

Researchers at Carnegie Mellon University have figured out a new method to create deepfakes -- videos that make it look like someone said or did something they didn't say or do. They're usually manipulated using machine learning and have been a big problem for porn sites, as they can be used to create fake celebrity pornographic videos or revenge porn. According to Gizmodo, the researchers "have figured out a way to automatically transfer the 'style' of one person to another." From the report: "For instance, Barack Obama's style can be transformed into Donald Trump," the researchers wrote in the description of a YouTube video highlighting the outcome of this method. The video shows the facial expressions of John Oliver transferred to both Stephen Colbert and an animated frog, from Martin Luther King, Jr. to Obama, and from Obama to Trump. The researchers describe the process in a paper as an "unsupervised data-driven approach." Like other methods of developing deepfakes, this one uses artificial intelligence. The paper doesn't exclusively deal in translating talking style and facial movements from one human to another -- it also includes examples with blooming flowers, sunrises and sunsets, and clouds and wind.

For the person-to-person deepfakes, the researchers cite examples of how certain mannerisms can be transferred, including "John Oliver's dimple while smiling, the shape of mouth characteristic of Donald Trump, and the facial mouth lines and smile of Stephen Colbert." The team used videos available to the public to develop these deepfakes. It's easy to see how these techniques might be applied in a more innocuous way...

An anonymous reader quotes a report from Motherboard: On Thursday, a group of researchers from Lancaster University posted a paper to arXiv that demonstrates how they used a smartphone's microphone and speaker system to steal the device's unlock pattern. Although the average person doesn't have to worry about getting hacked this way any time soon, the researchers are the first to demonstrate that this kind of attack is even possible. According to the researchers, their "SonarSnoop" attack decreases the number of unlock patterns an attacker must try by 70 percent and can be performed without the victim ever knowing they're being hacked. The attack begins when a user unwittingly installs a malicious application on their phone. When a user downloads the infected app, their phone begins broadcasting a sound signal that is just above the human range of hearing. This sound signal is reflected by every object around the phone, creating an echo. This echo is then recorded by the phone's microphone. By calculating the time between the emission of the sound and the return of its echo to the source, it is possible to determine the location of an object in a given space and whether that object is moving -- this is known as sonar.

The researchers were able to leverage this phenomenon to track the movement of someone's finger across a smartphone screen by analyzing the echoes recorded through the device's microphone. There are nearly 400,000 possible unlock patterns on the 3x3 swipe grid on Android phones, but prior research has demonstrated that 20 percent of people use one of 12 common patterns. While testing SonarSnoop, the researchers only focused on these dozen unlock combinations. Ten volunteers were recruited for the study and were asked to draw each of the 12 patterns five different times on a custom app. The researchers then tried a variety of sonar analysis techniques to reconstruct the password based on the acoustic signatures emitted by the phone. The best analysis technique resulted in the algorithm only having to try 3.6 out of the 12 possible patterns on average before it correctly determined the pattern.

An anonymous reader quotes a report from MIT Technology Review: David Anderson at Rydberg Technologies in Ann Arbor, Michigan, and a couple of colleagues, have reinvented the antenna from scratch. Their new device works in an entirely different way from conventional antennas, using a laser to measure the way radio signals interact with certain types of atoms. The secret sauce in the new device is Rydberg atoms. These are cesium atoms in which the outer electrons are so excited that they orbit the nucleus at great distance. At these distances, the electrons' potential energy levels are extremely closely spaced, and this gives them special properties. Indeed, any small electric field can nudge them from one level to another. Radio waves consist of alternating electric fields that readily interact with any Rydberg atoms they come across. This makes them potential sensors.

But how to detect this interaction? A gas made of Rydberg atoms has another property that turns out to be useful -- it can be made transparent by a laser tuned to a specific frequency. This laser essentially saturates the gas's ability to absorb light, allowing another laser beam to pass through it. However, the critical frequency at which this happens depends crucially on the properties of the Rydberg atoms in the gas. When these atoms interact with radio waves, the critical frequency changes in response. That's the basis of the radio detection. Anderson and co create a gas of cesium atoms excited into Rydberg states. They then use a laser tuned to a specific frequency to make the gas transparent. Finally, they shine a second laser through the gas and measure how much light is absorbed, to see how the transparency varies with ambient radio waves. The signal from a simple light-sensitive photodiode then reveals the way the radio signals are frequency modulated or amplitude modulated. The atomic radio can detect a huge range of signals -- over four octaves from the C band to the Q band, or wavelengths from 2.5 to 15 centimeters. It also should be less insensitive to electromagnetic interference due to its lack of conventional radio circuitry. "The atomic radio wave receiver operates by direct real-time optical detection of the atomic response to AM and FM baseband signals, precluding the need for traditional de-modulation and signal-conditioning electronics," say Anderson and co.

New research has shown just how bad AI is at dealing with online trolls. From a report: Such systems struggle to automatically flag nudity and violence, don't understand text well enough to shoot down fake news and aren't effective at detecting abusive comments from trolls hiding behind their keyboards. A group of researchers from Aalto University and the University of Padua found this out when they tested seven state-of-the-art models used to detect hate speech. All of them failed to recognize foul language when subtle changes were made, according to a paper [PDF] on arXiv. Adversarial examples can be created automatically by using algorithms to misspell certain words, swap characters for numbers or add random spaces between words or attach innocuous words such as 'love' in sentences.

New submitter georgecarlyle76 writes: Amazon AI researchers have publicly released a dataset of almost 400,000 transliterated names, to aid the development of natural-language-understanding systems that can search across databases that use different scripts. They describe the dataset's creation in a paper [PDF] they're presenting at COLING, together with experiments using the dataset to train different types of machine learning models.

It sounds like a joke, but the idea actually makes sense: More bugs, not less, could theoretically make a system safer. From a report: Carefully scatter non-exploitable decoy bugs in software, and attackers will waste time and resources on trying to exploit them. The hope is that attackers will get bored, overwhelmed, or run out of time and patience before finding an actual vulnerability. Computer science researchers at NYU suggested this strategy in a study published August 2, and call these fake-vulnerabilities "chaff bugs." Brendan Dolan-Gavitt, assistant professor at NYU Tandon and one of the researcher on this study, told me in an email that they've been working on techniques to automatically put bugs into programs for the past few years as a way to test and evaluate different bug-finding systems. Once they had a way to fill a program with bugs, they started to wonder what else they could do with it. "I also have a lot of friends who write exploits for a living, so I know how much work there is in between finding a bug and coming up with a reliable exploit -- and it occurred to me that this was something we might be able to take advantage of," he said. "People who can write exploits are rare, and their time is expensive, so if you can figure out how to waste it you can potentially have a great deterrent effect." Brendan has previously suggested that adding bugs to experimental software code could help with ultimately winding up with programs that have fewer vulnerabilities.

Kotlin, which Google blessed last year as an alternative to Java for programming Android apps, has already made its way into almost 12 per cent of open source Android apps, and in so doing has elevated their code quality. From a report: So we're told by computer scientists Bruno Gois Mateus and Matias Martinez, affiliated with University of Valenciennes in France, who observed that Google at the end of 2017 said Kotlin had infiltrated more than 17 per cent of Android apps developed with its IDE, Android Studio 3.0. Kotlin is an open source statically typed programing language that targets the JVM, Android, JavaScript (transpiling to ES5.1) and native platforms (via LLVM). JetBrains, the company that created it, contends Kotlin is more concise and more type-safe than Java. It estimates that apps written in Kotlin require about 40 per cent less code than they would with Java. With fewer lines of code, in theory, one can expect fewer bugs. In a paper distributed through pre-print service ArXiv, "An Empirical Study on Quality of Android Applications written in Kotlin language," Mateus and Martinez describe how they gathered 925 apps from the open source F-Droid repository, measured the amount of Kotlin code in each, and analyzed the code for "smells" as an indicator of code quality.

An anonymous reader writes: "Scientists from the University of California, Riverside (UCR) have published details last week about a new Spectre-class attack that they call SpectreRSB," reports Bleeping Computer. "Just like all 'Spectre-class' attacks, SpectreRSB takes advantage of the process of speculative execution -- a feature found in all modern CPUs that has the role of improving performance by computing operations in advance and later discarding unneeded data. The difference from previous Spectre-like attacks is that SpectreRSB recovers data from the speculative execution process by attacking a different CPU component involved in this 'speculation' routine, namely the Return Stack Buffer (RSB)." In a research paper, academics say they've used SpectreRSB attacks to recover data belonging to other processes, and have even tricked the RSB into spilling SGX secrets. The attack works on Intel, AMD, and ARM processors, known to use RSB. The attack can also bypass all the mitigations put in place for the original Spectre/Meltdown flaws.

An anonymous reader quotes a report from MIT Technology Review: Is it possible to see the effects of the Fukushima nuclear disaster in California wines produced at the time? Today we get an answer, thanks to a study carried out by french pharmacologist Philippe Hubert and a couple of colleagues. "In January 2017, we came across a series of Californian wines (Cabernet Sauvignon) from vintage 2009 to 2012," say Hubert and company. This set of wines provides the perfect test. The Fukushima disaster occurred on March 11, 2011. Any wine made before that date should be free of the effects, while any dating from afterward could show them. The team began their study with the conventional measurement of cesium-137 levels in the unopened bottles. That showed levels to be indistinguishable from background noise.

But the team was able to carry out more-sensitive tests by opening the wine and reducing it to ash by evaporation. This involves heating the wine to 100 degrees Celsius for one hour and then increasing the temperature to 500 degrees Celsius for eight hours. In this way, a standard 750-milliliter bottle of wine produces around four grams of ashes. The ashes were then placed in a gamma ray detector to look for signs of cesium-137. Using this method, Hubert and his colleagues found measurable amounts of cesium-137 above background levels in the wine produced after 2011. "It seems there is an increase in activity in 2011 by a factor of two," conclude the team.

Researchers from Nvidia, MIT, and Aalto University are using artificial intelligence to reduce noise in photos. The team used 50,000 images from the ImageNet dataset to train its AI system for reconstructing photos, and the system is able to remove noise from an image even though it has never seen the image without noise. VentureBeat: Named Noise2Noise, the AI system was created using deep learning and draws its intelligence from 50,000 images from the ImageNet database. Each came as a clean, high-quality image without noise but was manipulated to add randomized noise. Computer-generated images and MRI scans were also used to train Noise2Noise. Denoising or noise reduction methods have been around for a long time now, but methods that utilize deep learning are a more recent phenomenon.

A new study by Oxford University's Future of Humanity Institute determined that it's quite likely humans are alone in the observable universe. Fortune reports: The study looked at the Fermi paradox -- the apparent discrepancy between the seeming likelihood of alien life, given the billions of stars similar to our sun, and the scant evidence that such life actually exists. The paradox was named after physicist Enrico Fermi, who famously asked his colleagues at Los Alamos, N.M.. "Where Is Everyone?"

The study authors then examined various hypotheses and equations used to resolve the Fermi paradox. The results weren't pretty: "Our main result is to show that proper treatment of scientific uncertainties dissolves the Fermi paradox by showing that it is not at all unlikely ex ante for us to be alone in the Milky Way, or in the observable universe. Our second result is to show that, taking account of observational bounds on the prevalence of other civilizations, our updated probabilities suggest that there is a substantial probability that we are alone." SpaceX CEO Elon Musk cited the study's conclusions as an "added impetus" for humanity to become a spacefaring civilization capable of extending life beyond Earth. He tweeted: "This is why we must preserve the light of consciousness by becoming a spacefaring civilization & extending life to other planets..."

Last January in northern Sweden, a German-led team of physicists loaded a curious machine onto an unmanned rocket. The payload, about as tall as a single-story apartment, was essentially a custom-made freezer -- a vacuum chamber, with a small chip and lasers within, that could cool single atoms near absolute zero.

It may sound like a bizarre experiment, but it is something physicists have been aching to do for years. They launched the rocket about 90 miles past the atmosphere's boundary of outer space, monitoring a livestream from a heated building nearby. Then, just 17 minutes later, they watched as the freezer plummeted back down to Earth, landing via parachute on snowy ground 40 miles from the launch site. Wired elaborates: See, the freezer that the Germans launched has the ability to make atoms clump together in a cloud-like blob called a Bose-Einstein condensate -- a phase of matter that exhibits some truly bizarre properties. It's delicate enough to respond to tiny fluctuations in gravity and electromagnetic fields, which means it could someday make for a super-precise sensor in space. But down on Earth, it tends to collapse in a matter of milliseconds because of gravity. So the blobs had to go to space. Since the late '90s, physicists have been developing machines that can autonomously assemble and control the blobs during spaceflight. With this rocket launch, they've succeeded. The group in Germany, led by physicist Ernst Rasel of University of Hannover, just released pictures of blobs they managed to create [PDF], as well as precise measurements of how they jiggled during their brief trip. "They've essentially laid the groundwork to show that you can actually do this, and it's not totally insane," says physicist Nathan Lundblad of Bates College.

An anonymous reader quotes a report from MIT Technology Review: [Stephen McAleer and colleagues from the University of California, Irvine] have pioneered a new kind of deep-learning technique, called "autodidactic iteration," that can teach itself to solve a Rubik's Cube with no human assistance. The trick that McAleer and co have mastered is to find a way for the machine to create its own system of rewards. Here's how it works. Given an unsolved cube, the machine must decide whether a specific move is an improvement on the existing configuration. To do this, it must be able to evaluate the move. Autodidactic iteration does this by starting with the finished cube and working backwards to find a configuration that is similar to the proposed move. This process is not perfect, but deep learning helps the system figure out which moves are generally better than others. Having been trained, the network then uses a standard search tree to hunt for suggested moves for each configuration.

The result is an algorithm that performs remarkably well. "Our algorithm is able to solve 100% of randomly scrambled cubes while achieving a median solve length of 30 moves -- less than or equal to solvers that employ human domain knowledge," say McAleer and co. That's interesting because it has implications for a variety of other tasks that deep learning has struggled with, including puzzles like Sokoban, games like Montezuma's Revenge, and problems like prime number factorization. The paper on the algorithm -- called DeepCube -- is available on Arxiv.

Artem Tashkinov writes: It needs more sigmas, but Fermilab boffins in America are carefully speculating that they may have seen evidence of a new fundamental particle: the sterile neutrino. The suggestion follows tests conducted by the MiniBooNE (Mini Booster Neutrino Experiment) instrument, located near Chicago. Its mission is to detect neutrino mass through their oscillations. In the Standard Model of physics, neutrinos, like all particles, are initially assumed to be massless, but some observations, like neutrino oscillation, suggest there's mass there. The experiment that possibly detected sterile neutrinos collected 15 years of data from its commissioning in 2002, and the results have only now reached pre-press outlet arXiv.

Over 15 years, MiniBooNE detected a few hundred more electron neutrinos than were predicted in Standard Model theory. The extra particles suggests there is a fourth, heavier flavor. The findings bring the MiniBooNE team tantalizingly close to a "result" -- it's a 4.8 sigma result, when "discovery" demands 5 sigma.

Scientists from the Southwest Research Institute suggest Pluto may be a comet, as opposed to a planet or dwarf planet. According to a study published in the journal Icarus, Pluto could be made up of billions of comets all mashed together. Smithsonian reports: Scientists had long believed the dwarf planet Pluto was formed the way planets come to be: they start as swirling dust that's gradually pulled together by gravity. But with the realization that Pluto was a Kuiper belt dwarf planet, researchers began speculating about the origins of the icy world. The researchers turned to Sputnik Planitia -- the western lobe of the massive heart-shaped icy expanse stamped on Pluto's side -- for the task. As Christopher Glein, lead author of the paper and researcher at the Southwest Research Institute, explains to [Popular Science editor Neel V. Patel], the researchers used the data from New Horizons on this icy expanse to estimate the amount of nitrogen on Pluto and the amount that's escaped from its atmosphere.

Glein explains the conclusions in a statement: "We found an intriguing consistency between the estimated amount of nitrogen inside the [Sputnik Planitia] glacier and the amount that would be expected if Pluto was formed by the agglomeration of roughly a billion comets or other Kuiper Belt objects similar in chemical composition to 67P, the comet explored by Rosetta." The report goes on to mention a few caveats. "For one, researchers aren't sure that comet 67P has an average comet composition," reports Smithsonian. "For another, New Horizons only captured information about Pluto at a specific point in time, which means nitrogen rates could have changed over the last billions of years. [T]here's also still the possibility Pluto formed from cold ices with a chemical composition to that of the sun."

Artem Tashkinov shares a report from The Register: DeepMind has taught artificially intelligent programs to play classic Atari computer games by making them watch YouTube videos. Exploration games like 1984's Montezuma's Revenge are particularly difficult for AI to crack, because it's not obvious where you should go, which items you need and in which order, and where you should use them. That makes defining rewards difficult without spelling out exactly how to play the thing, and thus defeating the point of the exercise. For example, Montezuma's Revenge requires the agent to direct a cowboy-hat-wearing character, known as Panama Joe, through a series of rooms and scenarios to reach a treasure chamber in a temple, where all the goodies are hidden. Pocketing a golden key, your first crucial item, takes about 100 steps, and is equivalent to 100^18 possible action sequences.

To educate their code, the researchers chose three YouTube gameplay videos for each of the three titles: Montezuma's Revenge, Pitfall, and Private Eye. Each game had its own agent, which had to map the actions and features of the title into a form it could understand. The team used two methods: temporal distance classification (TDC), and cross-modal temporal distance classification (CDC). The DeepMind code still relies on lots of small rewards, of a kind, although they are referred to as checkpoints. While playing the game, every sixteenth video frame of the agent's session is taken as a snapshot and compared to a frame in a fourth video of a human playing the same game. If the agent's game frame is close or matches the one in the human's video, it is rewarded. Over time, it imitates the way the game is played in the videos by carrying out a similar sequence of moves to match the checkpoint frame. In the end, the agent was able to exceed average human players and other RL algorithms: Rainbow, ApeX, and DQfD. The researchers documented their method in a paper this week. You can view the agent in action here.

Shaun Nichols, reporting for The Register: A group of German researchers have devised a method to thwart the VM security in AMD's server chips. Dubbed SEVered (PDF), the attack would potentially allow an attacker, or malicious admin who had access to the hypervisor, the ability to bypass AMD's Secure Encrypted Virtualization (SEV) protections.

The problem, say Fraunhofer AISEC researchers Mathias Morbitzer, Manuel Huber, Julian Horsch and Sascha Wessel, is that SEV, which is designed to isolate VMs from the prying eyes of the hypervisor, doesn't fully isolate and encrypt the VM data within the physical memory itself.

tedlistens shares a report from Fast Company: To make AI easier for humans to understand and trust, researchers at the [Elon Musk-backed] nonprofit research organization OpenAI have proposed training algorithms to not only classify data or make decisions, but to justify their decisions in debates with other AI programs in front of a human or AI judge. In an experiment described in their paper (PDF), the researchers set up a debate where two software agents work with a standard set of handwritten numerals, attempting to convince an automated judge that a particular image is one digit rather than another digit, by taking turns revealing one pixel of the digit at a time. One bot is programmed to tell the truth, while another is programmed to lie about what number is in the image, and they reveal pixels to support their contentions that the digit is, say, a five rather than a six.

The image classification task, where most of the image is invisible to the judge, is a sort of stand-in for complex problems where it wouldn't be possible for a human judge to analyze the entire dataset to judge bot performance. The judge would have to rely on the facets of the data highlighted by debating robots, the researchers say. "The goal here is to model situations where we have something that's beyond human scale," says Geoffrey Irving, a member of the AI safety team at OpenAI. "The best we can do there is replace something a human couldn't possibly do with something a human can't do because they're not seeing an image."

An anonymous reader quotes a report from Gizmodo: Back in 2012, a Reddit user posted a map claiming to show the longest straight line that could be traversed across the ocean without hitting land. Intrigued, a pair of computer scientists have developed an algorithm that corroborates the route, while also demonstrating the longest straight line that can be taken on land. The researchers, Rohan Chabukswar from United Technologies Research Center Ireland, and Kushal Mukherjee from IBM Research India, created the algorithm in response to a map posted by reddit user user kepleronlyknows, who goes by Patrick Anderson in real life. His map showed a long, 20,000 mile route extending from Pakistan through the southern tips of Africa and South America and finally ending in an epic trans-Pacific journey to Siberia. On a traditional 2D map, the path looks nothing like a straight line; but remember, the Earth is a sphere.

Anderson didn't provide any evidence for the map, or an explanation for how the route was calculated. In light of this, Chabukswar and Mukherjee embarked upon a project to figure out if the straight line route was indeed the longest, and to see if it was possible for a computer algorithm to solve the problem, both for straight line passages on water without hitting land or an ice sheet, and for a continuous straight line passage on land without hitting a major body of water. Their ensuing analysis was posted to the pre-print arXiv server earlier this month, and has yet to go through peer review. "There would be 233,280,000 great circles to consider to find the global optimum, and each great circle would have 21,600 individual points to process -- a staggering 5,038,848,000,000 points to verify," the researchers wrote in their study.

An anonymous reader shares an excerpt from a report via Ars Technica: Researchers are proposing an idea to make your computer bigger. They are suggesting an extreme and awesome form of co-processing. They want to turn your entire house into a co-processor using the local Wi-Fi signal. Why, you may be asking, do we even want to do this in the first place? The real answer is to see if we can. But the answer given to funding agencies is thermal management. In a modern processor, if all the transistors were working all the time, it would be impossible to keep the chip cool. Instead, portions of the chip are put to sleep, even if that might mean slowing up a computation. But if, like we do with video cards, we farm out a large portion of certain calculations to a separate device, we might be able to make better use of the available silicon.

So, how do you compute with Wi-Fi in your bedroom? The basic premise is that waves already perform computations as they mix with each other, it's just that those computations are random unless we make some effort to control them. When two waves overlap, we measure the combination of the two: the amplitude of one wave is added to the amplitude of the other. Depending on the history of the two waves, one may have a negative amplitude, while the other may have a positive amplitude, allowing for simple computation. The idea here is to control the path that each wave takes so that, when they're added together, they perform the exact computation that we want them to. The classic example is the Fourier transform. A Fourier transform takes an object and breaks it down into a set of waves. If these waves are added together, the object is rebuilt. You can see an example of this in the animation here.

Chloe Rose Stuart-Ulin sheds some light on the world of paid impersonators on dating apps like Tinder. Here's an excerpt from the report: Every morning I wake up to the same routine. I log into the Tinder account of a 45-year-old man from Texas -- a client. I flirt with every woman in his queue for 10 minutes, sending their photos and locations to a central database of potential "Opportunities." For every phone number I get, I make $1.75. I'm what's called a "Closer" for the online-dating service ViDA (Virtual Dating Assistants). Men and women (though mostly men) from all over the world pay this company to outsource the labor and tedium of online dating. The matches I speak to on behalf of the Texan man and other clients have no idea they're chatting with a professional.

It shouldn't come as a surprise that these ghostwriting services exist. Tinder alone produces more than 12 million matches a day, and if you're a heterosexual American, you now have a one in three chance of meeting your future husband or wife online. But as e-romance hits an all-time high, our daily dose of rejection, harassment, and heartbreak creeps upward, too. Once you mix in the vague rules of netiquette and a healthy fear of catfishing scams, it's easy to see why someone might want to outsource their online-dating profile to a pro, if only to keep themselves sane. But where does the digital social assistant end and the con artist begin?

yagoda shares a report from MIT Technology Review: Andreas Rossler at the Technical University of Munich in Germany and colleagues have developed a deep-learning system that can automatically spot face-swap videos. The new technique could help identify forged videos as they are posted to the web. But the work also has sting in the tail. The same deep-learning technique that can spot face-swap videos can also be used to improve the quality of face swaps in the first place -- and that could make them harder to detect. The new technique relies on a deep-learning algorithm that Rossler and co have trained to spot face swaps. These algorithms can only learn from huge annotated data sets of good examples, which simply have not existed until now. In semi-related news, the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) says it's "fighting back" against the dangers posed by new face-swapping technologies that have been used to digitally superimpose the faces of its members onto the bodies of porn stars.

"SAG-AFTRA has undertaken an exhaustive review of our collective bargaining options and legislative options to combat any and all uses of digital re-creations, not limited to deepfakes, that defame our members and inhibit their ability to protect their images, voices and performances from misappropriation. We are talking with our members' representatives, union allies, and with state and federal legislators about this issue right now and have legislation pending in New York and Louisiana that would address this directly in certain circumstances. We also are analyzing state laws in other jurisdictions, including California, to make sure protections are in place. To the degree that there are not sufficient protections in place, we will work to fix that..."

A machine learning method which has been widely used in face recognition and other image- and speech-recognition applications, has shown promise in helping astronomers analyze images of galaxies and understand how they form and evolve. From a report: In a new study, accepted for publication in Astrophysical Journal and available online [PDF], researchers used computer simulations of galaxy formation to train a deep learning algorithm, which then proved surprisingly good at analyzing images of galaxies from the Hubble Space Telescope. The researchers used output from the simulations to generate mock images of simulated galaxies as they would look in observations by the Hubble Space Telescope. The mock images were used to train the deep learning system to recognize three key phases of galaxy evolution previously identified in the simulations. The researchers then gave the system a large set of actual Hubble images to classify.

The results showed a remarkable level of consistency in the neural network's classifications of simulated and real galaxies. "We were not expecting it to be all that successful. I'm amazed at how powerful this is," said coauthor Joel Primack, professor emeritus of physics and a member of the Santa Cruz Institute for Particle Physics (SCIPP) at UC Santa Cruz. "We know the simulations have limitations, so we don't want to make too strong a claim. But we don't think this is just a lucky fluke."

Adam Frank, writing for The Atlantic: We're used to imagining extinct civilizations in terms of the sunken statues and subterranean ruins. These kinds of artifacts of previous societies are fine if you're only interested in timescales of a few thousands of years. But once you roll the clock back to tens of millions or hundreds of millions of years, things get more complicated.

When it comes to direct evidence of an industrial civilization -- things like cities, factories, and roads -- the geologic record doesn't go back past what's called the Quaternary period 2.6 million years ago. For example, the oldest large-scale stretch of ancient surface lies in the Negev Desert. It's "just" 1.8 million years old -- older surfaces are mostly visible in cross section via something like a cliff face or rock cuts. Go back much farther than the Quaternary and everything has been turned over and crushed to dust.

And, if we're going back this far, we're not talking about human civilizations anymore. Homo sapiens didn't make their appearance on the planet until just 300,000 years or so ago. [...] Given that all direct evidence would be long gone after many millions of years, what kinds of evidence might then still exist? The best way to answer this question is to figure out what evidence we'd leave behind if human civilization collapsed at its current stage of development. Mr. Frank, along with Gavin Schmidt, Director of the NASA Goddard Institute for Space Studies, have published their research on the subject [PDF].

An anonymous reader quotes a report from Ars Technica: Google researchers have developed a deep-learning system designed to help computers better identify and isolate individual voices within a noisy environment. As noted in a post on the company's Google Research Blog this week, a team within the tech giant attempted to replicate the cocktail party effect, or the human brain's ability to focus on one source of audio while filtering out others -- just as you would while talking to a friend at a party. Google's method uses an audio-visual model, so it is primarily focused on isolating voices in videos. The company posted a number of YouTube videos showing the tech in action.

The company says this tech works on videos with a single audio track and can isolate voices in a video algorithmically, depending on who's talking, or by having a user manually select the face of the person whose voice they want to hear. Google says the visual component here is key, as the tech watches for when a person's mouth is moving to better identify which voices to focus on at a given point and to create more accurate individual speech tracks for the length of a video. According to the blog post, the researchers developed this model by gathering 100,000 videos of "lectures and talks" on YouTube, extracting nearly 2,000 hours worth of segments from those videos featuring unobstructed speech, then mixing that audio to create a "synthetic cocktail party" with artificial background noise added. Google then trained the tech to split that mixed audio by reading the "face thumbnails" of people speaking in each video frame and a spectrogram of that video's soundtrack. The system is able to sort out which audio source belongs to which face at a given time and create separate speech tracks for each speaker. Whew.

From a report on The Register: If you want your computer to be really secure, disconnect its power cable. So says Mordechai Guri and his team of side-channel sleuths at the Ben-Gurion University of the Negev. The crew have penned a paper titled PowerHammer: Exfiltrating Data from Air-Gapped Computers through Power Lines that explains how attackers could install malware that regulates CPU utilisation and creates fluctuations in the current flow that could modulate and encode data. The variations would be "propagated through the power lines" to the outside world.

Depending on the attacker's approach, data could be exfiltrated at between 10 and 1,000 bits-per-second. The higher speed would work if attackers can get at the cable connected to the computer's power supply. The slower speed works if attackers can only access a building's electrical services panel. The PowerHammer malware spikes the CPU utilisation by choosing cores that aren't currently in use by user operations (to make it less noticeable). Guri and his pals use frequency shift keying to encode data onto the line.

In a just-released white paper, the Army describes how it's working to make a battlefield network of machines and humans a reality. The Next Web reports: "Most of such intelligent things will not be too dissimilar from the systems we see on today's battlefield, such as unattended ground sensors, guided missiles (especially the fire-and-forget variety) and of course the unmanned aerial systems (UAVs)," reads the paper. "They will likely include physical robots ranging from very small size (such as an insect-scale mobile sensors) to large vehicle that can carry troops and supplies. Some will fly, others will crawl or walk or ride."

The paper was authored by the Army's chief of the Network Science Division of the Army Research Laboratory, Dr. Alexander Kott. It outlines the need to develop systems to augment both machines and people in the real world with artificially intelligent agents to defend the network: "In addition to physical intelligent things, the battlefield -- or at least the cyber domain of the battlefield -- will be populated with disembodied, cyber robots. These will reside within various computers and networks, and will move and acts in the cyberspace."

Kott takes pains to underscore the fact that the AI powering U.S. war efforts will need to be resilient in ways that today's AI simply isn't. He states: "The intelligent things will have to constantly think about an intelligent adversary that strategizes to deceive and defeat them. Without this adversarial intelligence, the battle things will not survive long enough to be useful." Ultimately, aside from outlining what the future battlefield will look like, the paper's conclusion is either disappointing or a giant relief, depending on your agenda: "Clearly, it is far beyond the current state of AI to operate intelligently in such an environments and with such demands. In particular, Machine Learning -- an area that has seen a dramatic progress in the last decade -- must experience major advances in order to become relevant to the real battlefield."

A new research paper [PDF] from Princeton University has found that 90 percent of affiliate posts on YouTube and Pinterest aren't disclosed to users. From a report: Affiliate links are customized URLs that content publishers can include in their posts. They're essentially ads, and publishers receive money from companies when users click on them. In the US, the Federal Trade Commission (FTC) requires that content makers identify when they're being paid to post something, but despite that, influencers continue to skirt around disclosures. The FTC has previously sent out letters to influencers reminding them of the requirement to communicate paid relationships with brands to their followers. The paper from Princeton analyzed over 500,000 YouTube videos and 2.1 million unique pins on Pinterest. Of those, 0.67 percent, or 3,472 videos on YouTube, and 0.85 percent, or 18,237 pins, contained affiliate links.

Researchers have discovered flaws in Monero, a digital currency that boasts a high degree of anonymity, that could lead to the identification of users. From a report: Monero is designed to mix up any given Monero "coin" with other payments, so that anyone scouring Monero's blockchain can't link it to any particular identity or previous transaction from the same source. But in a recent paper, a team of researchers from a broad collection of institutions -- including Princeton, Carnegie Mellon, Boston University, MIT, and the University of Illinois at Urbana-Champaign -- point to flaws in that mixing that make it possible to nonetheless extract individual transactions.

That shouldn't just worry anyone trying to stealthily spend Monero today. It also means evidence of earlier not-quite-untraceable payments remain carved into Monero's blockchain for years to come, visible for any snoop that cares to look.

"Robot brains will challenge the fundamental assumptions of how we humans do things," argues Popular Mechanics, noting that age-old truism "that computers will always do literally, exactly what you tell them to." A paper recently published to ArXiv highlights just a handful of incredible and slightly terrifying ways that algorithms think... An AI project which pit programs against each other in games of five-in-a-row Tic-Tac-Toe on an infinitely expansive board surfaced the extremely successful method of requesting moves involving extremely long memory addresses which would crash the opponent's computer and award a win by default...

These amusing stories also reflect the potential for evolutionary algorithms or neural networks to stumble upon solutions to problems that are outside-the-box in dangerous ways. They're a funnier version of the classic AI nightmare where computers tasked with creating peace on Earth decide the most efficient solution is to exterminate the human race. The solution, the paper suggests, is not fear but careful experimentation.
The paper (available as a free download) contains 27 anecdotes, which its authors describe as a "crowd-sourced product of researchers in the fields of artificial life and evolutionary computation. Popular Science adds that "the most amusing examples are clearly ones where algorithms abused bugs in their simulations -- essentially glitches in the Matrix that gave them superpowers."

"Robot brains will challenge the fundamental assumptions of how we humans do things," argues Popular Mechanics, noting that age-old truism "that computers will always do literally, exactly what you tell them to." A paper recently published to ArXiv highlights just a handful of incredible and slightly terrifying ways that algorithms think... An AI project which pit programs against each other in games of five-in-a-row Tic-Tac-Toe on an infinitely expansive board surfaced the extremely successful method of requesting moves involving extremely long memory addresses which would crash the opponent's computer and award a win by default...

These amusing stories also reflect the potential for evolutionary algorithms or neural networks to stumble upon solutions to problems that are outside-the-box in dangerous ways. They're a funnier version of the classic AI nightmare where computers tasked with creating peace on Earth decide the most efficient solution is to exterminate the human race. The solution, the paper suggests, is not fear but careful experimentation.
The paper (available as a free download) contains 27 anecdotes, which its authors describe as a "crowd-sourced product of researchers in the fields of artificial life and evolutionary computation. Popular Science adds that "the most amusing examples are clearly ones where algorithms abused bugs in their simulations -- essentially glitches in the Matrix that gave them superpowers."

Slashdot reader Chris Reeve writes: An October 2017 paper titled Electric Currents along Astrophysical Jets reports that "Several researchers have reported direct evidence for large scale electric currents along astrophysical jets." A review of the citations at the end of that paper and others (here and here, for instance) would seem to suggest that one of the great Internet science debates has finally been settled: Electricity does indeed travel through space over vast cosmic distances.

What has been interesting to watch about this unexpected development is that science journalists have so far not explicitly reported this as a shift in theory, and commenters on sites like phys.org appear to deny that any change has even occurred: "The jets have been shown not to be electric currents, the energy and the physics involved are certainly not electromagnetic." This comment completely rejecting these new findings was highly rated by other phys.org readers, suggesting that the failure to explicitly report this as a change in theory has left this controversial topic in a highly confused state.
The paper summarizes what it calls "observational evidence for the existence of large scale electric currents and their associated grand design helical magnetic fields in kpc-scale astrophysical jets." And the original submitter details the history of the question in a follow-up comment arguing that at our current moment in time, "a mistaken bias against electricity in space continues to dominate conversations."

An anonymous reader quotes a report from Tom's Hardware: Google announced a 72-qubit universal quantum computer that promises the same low error rates the company saw in its first 9-qubit quantum computer. Google believes that this quantum computer, called Bristlecone, will be able to bring us to an age of quantum supremacy. In a recent announcement, Google said: "If a quantum processor can be operated with low enough error, it would be able to outperform a classical supercomputer on a well-defined computer science problem, an achievement known as quantum supremacy. These random circuits must be large in both number of qubits as well as computational length (depth). Although no one has achieved this goal yet, we calculate quantum supremacy can be comfortably demonstrated with 49 qubits, a circuit depth exceeding 40, and a two-qubit error below 0.5%. We believe the experimental demonstration of a quantum processor outperforming a supercomputer would be a watershed moment for our field, and remains one of our key objectives."

According to Google, a minimum error rate for quantum computers needs to be in the range of less than 1%, coupled with close to 100 qubits. Google seems to have achieved this so far with 72-qubit Bristlecone and its 1% error rate for readout, 0.1% for single-qubit gates, and 0.6% for two-qubit gates. Quantum computers will begin to become highly useful in solving real-world problems when we can achieve error rates of 0.1-1% coupled with hundreds of thousand to millions of qubits. According to Google, an ideal quantum computer would have at least hundreds of millions of qubits and an error rate lower than 0.01%. That may take several decades to achieve, even if we assume a "Moore's Law" of some kind for quantum computers (which so far seems to exist, seeing the progress of both Google and IBM in the past few years, as well as D-Wave).

An anonymous reader shares a report: AI research and video games are a match made in heaven. Researchers get a ready-made virtual environment with predefined goals they can control completely, and the AI agent gets to romp around without doing any damage. Sometimes, though, they do break things. Case in point is a paper published this week by a trio of machine learning researchers from the University of Freiburg in Germany. They were exploring a particular method of teaching AI agents to navigate video games (in this case, desktop ports of old Atari titles from the 1980s) when they discovered something odd. The software they were testing discovered a bug in the port of the retro video game Q*bert that allowed it to rack up near infinite points. As the trio describe in the paper, published on pre-print server arXiv, the agent was learning how to play Q*bert when it discovered an "interesting solution." Normally, in Q*bert, players jump from cube to cube, with this action changing the platforms' colors. Change all the colors (and dispatch some enemies), and you're rewarded with points and sent to the next level. The AI found a better way, though: "First, it completes the first level and then starts to jump from platform to platform in what seems to be a random manner. For a reason unknown to us, the game does not advance to the second round but the platforms start to blink and the agent quickly gains a huge amount of points (close to 1 million for our episode time limit)."

Scientists are worried that space aliens might send messages that worm their way into human society -- not to steal our passwords but to bring down our culture. "Astrophysicists Michael Hippke and John Learned argue in a recent paper that our telescopes might pick up hazardous messages sent our way -- a virus that shuts down our computers, for example, or something a bit like cosmic blackmail: 'Do this for us, or we'll make your sun go supernova and destroy Earth,'" reports NBC News. "Or perhaps the cosmic hackers could trick us into building self-replicating nanobots, and then arrange for them to be let loose to chew up our planet or its inhabitants." From the report: The astrophysicists also suggest that the extraterrestrials could show their displeasure (what did we do?) by launching a cyberattack. Maybe you've seen the 1996 film "Independence Day," in which odious aliens are vanquished by a computer virus uploaded into their machinery. That's about as realistic as sabotaging your neighbor's new laptop by feeding it programs written for the Commodore 64. In other words, aliens that could muster the transmitter power (not to mention the budget) to try wiping us out with code are going to have a real compatibility problem.

Yet there is a way that messages from space might be disruptive. Extraterrestrials could simply give us some advanced knowledge -- not as a trade, but as a gift. How could that possibly be a downer? Imagine: You're a physicist who has dedicated your career to understanding the fundamental structure of matter. You have a stack of reprints, a decent position, and a modicum of admiration from the three other specialists who have read your papers. Suddenly, aliens weigh in with knowledge that's a thousand years ahead of yours. So much for your job and your sense of purpose. If humanity is deprived of the opportunity to learn things on its own, much of its impetus for novelty might evaporate. In a society where invention and discovery are written out of the script, progress and improvement would suffer.

New submitter the gmr writes: According to an announcement on the OpenAI blog, Elon Musk has stepped down from the board of directors of the nonprofit AI safety group, which he co-founded in 2015, due potential conflict of interest with his company Tesla. As explained in a post on Futurism, the move away from OpenAI may indicate that Tesla may be moving forward with more AI projects than most people may realize. Musk's departure may mean that Tesla is closer to delivering vehicles capable of Level 5 autonomy, "fully self-driving" vehicles that more than 35,000 Tesla customers paid for even though the technology does not yet exist. "Elon Musk will depart the OpenAI Board but will continue to donate and advise the organization," the announcement reads. "As Tesla continues to become more focused on AI, this will eliminate a potential future conflict for Elon." The OpenAI board of directors now consists of Greg Brockman, Ilya Sutskever, Holden Karnofsky, and Sam Altman, with whom Musk co-founded the venture. The company reportedly plans to not only fill Musk's seat but expand their team as well.

"Open AI has also been a prominent voice in the conversation concerning the limitations, challenges, and potential dangers of artificial intelligence," reports Futurism. "Just this week, the company co-released a report with a number of other global AI experts that outlines the potential 'malicious' uses of the technology and how to prevent them."

The Register: A team within Google Brain -- the web giant's crack machine-learning research lab -- has taught software to generate Wikipedia-style articles by summarizing information on web pages... to varying degrees of success. As we all know, the internet is a never ending pile of articles, social media posts, memes, joy, hate, and blogs. It's impossible to read and keep up with everything. Using AI to tell pictures of dogs and cats apart is cute and all, but if such computers could condense information down into useful snippets, that would be really be handy. It's not easy, though. A paper, out last month and just accepted for this year's International Conference on Learning Representations (ICLR) in April, describes just how difficult text summarization really is. A few companies have had a crack at it. Salesforce trained a recurrent neural network with reinforcement learning to take information and retell it in a nutshell, and the results weren't bad.

From a research paper on Arxiv: The problem of detecting bots, automated social media accounts governed by software but disguising as human users, has strong implications. For example, bots have been used to sway political elections by distorting online discourse, to manipulate the stock market, or to push anti-vaccine conspiracy theories that caused health epidemics. Most techniques proposed to date detect bots at the account level, by processing large amount of social media posts, and leveraging information from network structure, temporal dynamics, sentiment analysis, etc. In this paper [PDF], we propose a deep neural network based on contextual long short-term memory (LSTM) architecture that exploits both content and metadata to detect bots at the tweet level: contextual features are extracted from user metadata and fed as auxiliary input to LSTM deep nets processing the tweet text.

From a research paper on Arxiv: The problem of detecting bots, automated social media accounts governed by software but disguising as human users, has strong implications. For example, bots have been used to sway political elections by distorting online discourse, to manipulate the stock market, or to push anti-vaccine conspiracy theories that caused health epidemics. Most techniques proposed to date detect bots at the account level, by processing large amount of social media posts, and leveraging information from network structure, temporal dynamics, sentiment analysis, etc. In this paper [PDF], we propose a deep neural network based on contextual long short-term memory (LSTM) architecture that exploits both content and metadata to detect bots at the tweet level: contextual features are extracted from user metadata and fed as auxiliary input to LSTM deep nets processing the tweet text.

sciencehabit shares a report from Science Magazine: SpaceX CEO Elon Musk grabbed the world's attention last week after launching his Tesla Roadster into space. But his publicity stunt has a half-life way beyond even what he could imagine -- the Roadster should continue to orbit through the solar system, perhaps slightly battered by micrometeorites, for a few tens of millions of years. Now, a group of researchers specializing in orbital dynamics has analyzed the car's orbit for the next few million years. And although it's impossible to map it out precisely, there is a small chance that one day it could return and crash into Earth. But don't panic: That chance is just 6% over a million years, and it would likely burn up as it entered the atmosphere.

Hanno Rein of the University of Toronto in Canada and his colleagues regularly model the motions of planets and exoplanets. "We have all the software ready, and when we saw the launch last week we thought, 'Let's see what happens.' So we ran the [Tesla's] orbit forward for several million years," he says. The Falcon Heavy rocket from SpaceX propelled the car out toward Mars, but the sun's gravity will bring it swinging in again some months from now in an elliptical orbit, so it will repeatedly cross the orbits of Mars, Earth, and Venus until it sustains a fatal accident. The Roadster's first close encounter with Earth will be in 2091 -- the first of many in the millennia to come.

jwhyche, Slashdot reader #6,192, writes: If you bought some illegal narcotics off Silk Road or even gave money to Wikileaks. Researchers at Qatar University and Hamad Bin Khalifa University have been able to link these transactions with real world identities. They have been able to do this even if the transactions are years old. Their research shows how easy it is to link accounts to these transactions without using any of the tools available to law enforcement like search warrants or subpoenas.
The researchers started with 88 unique bitcoin addresses from Tor hidden services, and then searched 5 billion tweets and 1 million pages on the Bitcoin Talk forum -- ultimately linking 125 unique users to 20 Tor hidden services. "Bitcoin addresses should always be considered exploitable," the researchers conclude, "as they can be used to deanonymize users retroactively."

Their paper is titled "When a Small Leak Sinks a Great Ship: Deanonymizing Tor Hidden Service Users Through Bitcoin Transactions Analysis," and Wired summarizes one of their conclusions. "Even deleting profile information that includes bitcoin addresses may not be enough if a post has been cached or captured by services like the Internet Archive, they point out. 'If you're vulnerable now, you're vulnerable in the future.'"

"Researchers at Google were able to create little stickers with 'psychedelic'-looking patterns on them that could trick computer AI image-classifying algorithms into mis-classifying images of objects that it would normally be able to recognize," writes amxcoder: The patterned stickers work by tricking the image recognition algorithm into focusing on, and studying, the little pattern on the small sticker -- and ignoring the rest of the image, including the actual object in the picture... The images on the stickers were created by the researchers using knowledge of features and shapes, patterns, and colors that the image recognition algorithms look for and focus on.

These stickers were created so that the algorithm finds them 'more interesting' than the rest of the image and will focus most of it's attention on analyzing the pattern, while giving the rest of the image content a lower importance, thus ignoring it or confusing it.
The technique "works in the real world, and can be disguised as an innocuous sticker," note the researchers -- describing them as "targeted adversarial image patches."

An analysis by more than 200 astronomers has been published that shows the mysterious dimming of star KIC 8462852 -- nicknamed Tabby's star -- is not being produced by an alien megastructure. "The evidence points most strongly to a giant cloud of dust occasionally obscuring the star," reports The Guardian. From the report: KIC 8462852 is approximately 1,500 light years away from the Earth and hit the headlines in October 2015 when data from Nasa's Kepler space telescope showed that it was dimming by unexplainably large amounts. The star's light dropped by 20% first and then 15% making it unique. Even a large planet passing in front of the star would have blocked only about 1% of the light. For an object to block 15-20%, it would have to be approaching half the diameter of the star itself. With this realization, a few astronomers began whispering that such a signal would be the kind expected from a gigantic extraterrestrial construction orbiting in front of the star -- and the idea of the alien megastructure was born.

In the case of Tabby's star, the new observations show that it dims more at blue wavelengths than red. Thus, its light is passing through a dust cloud, not being blocked by an alien megastructure in orbit around the star. The new analysis of KIC 8462852 showing these results is to be published in The Astrophysical Journal Letters. It reinforces the conclusions reached by Huan Meng, University of Arizona, Tucson, and collaborators in October 2017. They monitored the star at multiple wavelengths using Nasa's Spitzer and Swift missions, and the Belgian AstroLAB IRIS observatory. These results were published in The Astrophysical Journal.

Catalin Cimpanu, writing for BleepingComputer: Attackers can use sound waves to interfere with a hard drive's normal mode of operation, creating a temporary or permanent denial of state (DoS) that could be used to prevent CCTV systems from recording video footage or freeze computers dealing with critical operations. The basic principle behind this attack is that sound waves introduce mechanical vibrations into an HDD's data-storage platters. If the sound is played at a specific frequency, it creates a resonance effect that amplifies the vibration effect. Because hard drives store vasts amounts of information inside small areas of each platter, they are programmed to stop all read/write operations during the time a platter vibrates so to avoid scratching storage disks and permanently damaging an HDD. Last week, scientists from the Princeton and Purdue universities published new research into the topic, expanding on the previous findings with the results of additional practical tests. The research team used a specially crafted test rig to blast audio waves at a hard drive from different angles, recording results to determine the sound frequency, attack time, distance from the hard drive, and sound wave angle at which the HDD stopped working.

An anonymous reader quotes a report from Quartz: A research paper published by Google this month -- which has not been peer reviewed -- details a text-to-speech system called Tacotron 2, which claims near-human accuracy at imitating audio of a person speaking from text. The system is Google's second official generation of the technology, which consists of two deep neural networks. The first network translates the text into a spectrogram (pdf), a visual way to represent audio frequencies over time. That spectrogram is then fed into WaveNet, a system from Alphabet's AI research lab DeepMind, which reads the chart and generates the corresponding audio elements accordingly. The Google researchers also demonstrate that Tacotron 2 can handle hard-to-pronounce words and names, as well as alter the way it enunciates based on punctuation. For instance, capitalized words are stressed, as someone would do when indicating that specific word is an important part of a sentence. Quartz has embedded several different examples in their report that feature a sentence generated by AI along with a sentence read aloud from a human hired by Google. Can you tell which is the AI generated sample?

Science_afficionado writes: By measuring the flicker pattern of light from distant stars, astronomers have developed a new and improved method for measuring the masses of millions of solitary stars, especially those hosting exoplanets. Stevenson Professor of Physics and Astronomy Keivan Stassun says, "First, we use the total light from the star and its parallax to infer its diameter. Next, we analyze the way in which the light from the star flickers, which provides us with a measure of its surface gravity. Then we combine the two to get the star's total mass." Stassun and his colleagues describe the method and demonstrate its accuracy using 675 stars of known mass in an article titled "Empirical, accurate masses and radii of single stars with TESS and GAIA" accepted for publication in the Astronomical Journal.

David Salisbury via Vanderbilt University explains the other methods of determining the mass of distant stars, and why they aren't always the most accurate: "Traditionally, the most accurate method for determining the mass of distant stars is to measure the orbits of double star systems, called binaries. Newton's laws of motion allow astronomers to calculate the masses of both stars by measuring their orbits with considerable accuracy. However, fewer than half of the star systems in the galaxy are binaries, and binaries make up only about one-fifth of red dwarf stars that have become prized hunting grounds for exoplanets, so astronomers have come up with a variety of other methods for estimating the masses of solitary stars. The photometric method that classifies stars by color and brightness is the most general, but it isn't very accurate. Asteroseismology, which measures light fluctuations caused by sound pulses that travel through a star's interior, is highly accurate but only works on several thousand of the closest, brightest stars." Stassun says his method "can measure the mass of a large number of stars with an accuracy of 10 to 25 percent," which is "far more accurate than is possible with other available methods, and importantly it can be applied to solitary stars so we aren't limited to binaries."

An anonymous reader quotes a report from The Verge: In a new paper published this week, DeepMind describes how a descendant of the AI program that first conquered the board game Go has taught itself to play a number of other games at a superhuman level. After eight hours of self-play, the program bested the AI that first beat the human world Go champion; and after four hours of training, it beat the current world champion chess-playing program, Stockfish. Then for a victory lap, it trained for just two hours and polished off one of the world's best shogi-playing programs named Elmo (shogi being a Japanese version of chess that's played on a bigger board). One of the key advances here is that the new AI program, named AlphaZero, wasn't specifically designed to play any of these games. In each case, it was given some basic rules (like how knights move in chess, and so on) but was programmed with no other strategies or tactics. It simply got better by playing itself over and over again at an accelerated pace -- a method of training AI known as "reinforcement learning."

schwit1 quotes ScienceAlert: In May 2017, researchers at Google Brain announced the creation of AutoML, an artificial intelligence (AI) that's capable of generating its own AIs. More recently, they decided to present AutoML with its biggest challenge to date, and the AI that can build AI created a 'child' that outperformed all of its human-made counterparts... For this particular child AI, which the researchers called NASNet, the task was recognising objects -- people, cars, traffic lights, handbags, backpacks, etc. -- in a video in real-time. AutoML would evaluate NASNet's performance and use that information to improve its child AI, repeating the process thousands of times.

When tested on the ImageNet image classification and COCO object detection data sets NASNet was 82.7 percent accurate at predicting images on ImageNet's validation set. This is 1.2 percent better than any previously published results, and the system is also 4 percent more efficient, with a 43.1 percent mean Average Precision (mAP).

sciencehabit shares a report from Science Magazine: Automatic language translation has come a long way, thanks to neural networks -- computer algorithms that take inspiration from the human brain. But training such networks requires an enormous amount of data: millions of sentence-by-sentence translations to demonstrate how a human would do it. Now, two new papers show that neural networks can learn to translate with no parallel texts -- a surprising advance that could make documents in many languages more accessible.

The two new papers, both of which have been submitted to next year's International Conference on Learning Representations but have not been peer reviewed, focus on another method: unsupervised machine learning. To start, each constructs bilingual dictionaries without the aid of a human teacher telling them when their guesses are right. That's possible because languages have strong similarities in the ways words cluster around one another. The words for table and chair, for example, are frequently used together in all languages. So if a computer maps out these co-occurrences like a giant road atlas with words for cities, the maps for different languages will resemble each other, just with different names. A computer can then figure out the best way to overlay one atlas on another. Voila! You have a bilingual dictionary. The studies -- "Unsupervised Machine Translation Using Monolingual Corpora Only" and "Unsupervised Neural Machine Translation" -- were both submitted to the e-print archive arXiv.org.