Xiaomi spin-off brand Poco has launched its successor to last year's head-turning Poco F1. The Poco X2, as it is called, is a high-performance, low-cost phone for the extremely competitive India market. It'll be available in India on February 11th; it's unclear if the phone will be released elsewhere. The Verge has the details: The X2 has a Snapdragon 730G processor, a 6.67-inch 20:9 1080p LCD screen with a 120Hz refresh rate, and a 4,500mAh battery with 27W fast charging. The phone charges over USB-C, includes a headphone jack, and has a fingerprint sensor integrated into the power button. It's the first phone in India to feature Sony's new 64-megapixel IMX686 sensor, the successor to the 48-megapixel IMX586 that dominated the phone landscape in 2019, and that part is backed by an 8-megapixel ultrawide camera, a 2-megapixel macro module, and a 2-megapixel depth sensor. There's also a 20-megapixel selfie camera with a 2-megapixel depth sensor housed in a hole-punch display cutout.

The director of astrobiology at Columbia University saw something this week that he just had to respond to: Elon Musk "talking about sending 1 million people to Mars by 2050, using no less than three Starship launches per day (with a stash of 1,000 of these massive spacecraft on call)."

A reader shared this article from Scientific American: The martian radiation environment is a problem for human explorers that cannot be overstated... For reasons unclear to me, this tends to get pushed aside compared to other questions to do with Mars's atmosphere (akin to sitting 30km [18.6 miles] above Earth with no oxygen), temperatures, natural resources (water), nasty surface chemistry (perchlorates), and lower surface gravitational acceleration (1/3rd that on Earth). But we actually have rather good data on the radiation situation on Mars (and in transit to Mars) from the Radiation Assessment Detector (RAD) that has been riding along with the Curiosity rover since its launch from Earth.

The bottom line is that the extremely thin atmosphere on Mars, and the absence of a strong global magnetic field, result in a complex and potent particle radiation environment. There are lower energy solar wind particles (like protons and helium nuclei) and much higher energy cosmic ray particles crashing into Mars all the time. The cosmic rays, for example, also generate substantial secondary radiation -- crunching into martian regolith to a depth of several meters before hitting an atomic nucleus in the soil and producing gamma-rays and neutron radation... [I]f we consider just the dose on Mars, the rate of exposure averaged over one Earth year is just over 20 times that of the maximum allowed for a Department of Energy radiation worker in the US (based off of annual exposure)....

[Y]ou'd need to put a few meters of regolith above you, or live in some deep caves and lava tubes to dodge the worst of the radiation. And then there are risks not to do with cancer that we're only just beginning to learn about. Specifically, there is evidence that neurological function is particularly sensitive to radiation exposure, and there is the question of our essential microbiome and how it copes with long-term, persistent radiation damage. Finally, as Hassler et al. discuss, the "flavor" (for want of a better word) of the radiation environment on Mars is simply unlike that on Earth, not just measured by extremes but by its make up, comprising different components than on Earth's surface.

To put all of this another way: in the worst case scenario (which may or may not be a realistic extrapolation) there's a chance you'd end up dead or stupid on Mars. Or both.

How an overlooked feature of deep learning networks can turn into a major breakthrough for AI. From a report: Traditionally, deep learning algorithms such as deep neural networks (DNNs) are trained in a supervised fashion to recognize specific classes of things. In a typical task, a DNN might be trained to visually recognize a certain number of classes, say pictures of apples and bananas. Deep learning algorithms, when fed a good quantity and quality of data, are really good at coming up with precise, low error, confident classifications. The problem arises when a third, unknown object appears in front of the DNN. If an unknown object that was not present in the training set is introduced, such as an orange, then the network will be forced to "guess" and classify the orange as the closest class that captures the unknown object -- an apple! Basically, the world for a DNN trained on apples and bananas is completely made of apples and bananas. It can't conceive the whole fruit basket.

While its usefulness is not immediately clear in all applications, the idea of "nothing" or a "class zero" is extremely useful in several ways when training and deploying a DNN. During the training process, if a DNN has the ability to classify items as "apple," "banana," or "nothing," the algorithm's developers can determine if it hasn't effectively learned to recognize a particular class. That said, if pictures of fruit continue to yield "nothing" responses, perhaps the developers need to add another "class" of fruit to identify, such as oranges. Meanwhile, in a deployment scenario, a DNN trained to recognize healthy apples and bananas can answer "nothing" if there is a deviation from the prototypical fruit it has learned to recognize. In this sense, the DNN may act as an anomaly detection network -- aside from classifying apples and bananas, it can also, without further changes, signal when it sees something that deviates from the norm. As of today, there are no easy ways to train a standard DNN so that it can provide the functionality above.

An anonymous reader quotes a report from Scientific American: A research group at the medical devices and health care giant Abbott has discovered a new strain of human immunodeficiency virus, or HIV -- the first to be identified in 19 years. Abbott continues to look for potential new HIV strains to ensure that diagnostic tests for blood screening and detecting infectious diseases remain up to date, says Mary Rodgers, senior author of the paper announcing the finding and head of the company's Global Viral Surveillance Program. The new strain, called HIV-1 group M subtype L, is extremely rare and can be detected by Abbott's current screening system, Rodgers says. The company's tests screen more than 60 percent of the global blood supply, she adds, noting it must detect every strain and "has to be right every time."

The most recent of the three samples used to identify HIV-1 group M subtype L has been sitting in an Abbott freezer since 2001. The amount of virus in the sample was too low to read back then, but new technology recently made it possible. Comparing that sequence with the others made available by the research community, Abbott researchers found two additional examples of the strain -- in samples from 1983 and 1990, also from the Democratic Republic of the Congo, hinting that it has been around for a while. "Now that we know it exists, it'll change how we look for it," Rodgers says. The company's tests focus on the part of the viral genome that does not change very much from generation to generation, which is why it was able to detect the new strain. The finding also suggests there are more strains to be found, Rodgers says. "The full diversity has not been characterized. We're going to continue to look." The study has been published in the Journal of Acquired Immune Deficiency Syndromes.

As consumers, we all have "secret scores": hidden ratings that determine how long each of us waits on hold when calling a business, whether we can return items at a store, and what type of service we receive. A low score sends you to the back of the queue; high scores get you elite treatment. From a report: Every so often, journalists lament these systems' inaccessibility. They're "largely invisible to the public," The New York Times wrote in 2012. "Most people have no inkling they even exist," The Wall Street Journal said in 2018. Most recently, in April, The Journal's Christopher Mims looked at a company called Sift, whose proprietary scoring system tracks 16,000 factors for companies like Airbnb and OkCupid. "Sift judges whether or not you can be trusted," he wrote, "yet there's no file with your name that it can produce upon request." As of this summer, though, Sift does have a file on you, which it can produce upon request. I got mine, and I found it shocking: More than 400 pages long, it contained all the messages I'd ever sent to hosts on Airbnb; years of Yelp delivery orders; a log of every time I'd opened the Coinbase app on my iPhone. Many entries included detailed information about the device I used to do these things, including my IP address at the time.

Sift knew, for example, that I'd used my iPhone to order chicken tikka masala, vegetable samosas and garlic naan on a Saturday night in April three years ago. It knew I used my Apple laptop to sign into Coinbase in January 2017 to change my password. Sift knew about a nightmare Thanksgiving I had in California's wine country, as captured in my messages to the Airbnb host of a rental called "Cloud 9." This may sound somewhat comical, but the companies gathering and paying for this data find it extremely valuable for rooting out fraud and increasing the revenue they can collect from big spenders. Sift has this data because the company has been hired by Airbnb, Yelp, and Coinbase to identify stolen credit cards and help spot identity thieves and abusive behavior. Still, the fact that obscure companies are accumulating information about years of our online and offline behavior is unsettling, and at a minimum it creates the potential for abuse or discrimination -- particularly when those companies decide we don't stack up.

An anonymous reader quotes a report from Fast Company: Motorola Solutions (not to be confused with Motorola Mobility, which makes the smartphones you know), the biggest global player in these LMR walkies, is releasing what appears to be the most advanced walkie-talkie ever. Called the APX Next, it's a chunky black brick with a thick antenna and a giant push-to-talk button on its side. Much like an iPhone, it also features a touchscreen on its front -- but don't be distracted by that. Its real innovation was born from 2,000 hours of interviews and testing with more than 50 emergency service agencies, including SWAT teams and detectives. It's a voice-recognition system that can operate in extremely loud environments, with artificial intelligence software that can look up 95 of the most common things a police officer or firefighter would call into dispatch -- like a driver's license, or license plate -- without any human operator on the other end of the line. But its ultimate promise is simply to free up the user's hands as much as possible, ensuring that someone is as safe and capable as possible during an emergency.

APX Next is a walkie-talkie and a cellphone combined. It has both the high-powered radio chip for land communications and a low-powered 4G/LTE chip for cell-tower data. These two chips can work at the same time, which is an engineering challenge, especially because the walkie-radio has 25 times the wattage of the 4G chip. The core buttons, including the large talk button, are all designed as you'd expect, to ensure they can be used without looking, and purely by muscle memory in stressful situations. Four separate microphones capture your voice, with programming designed specifically to cancel out exceptionally loud noises. But it isn't always listening for a wake word like the Echo or Google Home. You need to hit a button to cue the assistant. The company says it's using unnamed third parties to handle the natural language processing in the cloud. "What Motorola did was train the model specifically to handle things like ten-codes and even regionally specific dialects across the U.S," the report says.

"Once a question is sent to the cloud, the AI is able to scour a city or force's database for the same private information a dispatcher would be looking up." What's also neat is that the AI won't automatically read sensitive information out loud. Instead, the radio will beep when it has an answer, and the user can get to a private place, if they wish, before hitting a button to hear the results.

Apple's Deep Fusion photography system has arrived in the latest developer betas of iOS 13, hopefully hinting that it will ship for the iPhone 11 and 11 Pro soon. From a report: To refresh your memory, Deep Fusion is a new image processing pipeline for medium-light images, which Apple senior VP Phil Schiller called "computational photography mad science" when he introduced it onstage. But like much of iOS 13, Deep Fusion wasn't ready when the phones arrived two weeks ago. And although the iPhone 11 and 11 Pro have extremely impressive cameras, Deep Fusion's is meant to offer a massive step forward in indoor and medium-lighting situations. And since so many photos are taken indoors and in medium light, we're looking forward to testing it.

[...] With Deep Fusion, the iPhone 11 and 11 Pro cameras will have three modes of operation that automatically kick in based on light levels and the lens you're using: The standard wide angle lens will use Apple's enhanced Smart HDR for bright to medium-light scenes, with Deep Fusion kicking in for medium to low light, and Night mode coming on for dark scenes. The tele lens will mostly use Deep Fusion, with Smart HDR only taking over for very bright scenes, and Night mode for very dark scenes. The ultrawide will always use Smart HDR, as it does not support either Deep Fusion or Night mode.

An anonymous reader quotes a report from Ars Technica: Sega and Sports Interactive have announced that Football Manager 2020 will be sold in new eco-friendly package that uses much less plastic, and they're pushing for the rest of the entertainment industry to follow suit. The new packaging replaces the now-standard plastic DVD case used for most game discs with a folded, reinforced cardboard sleeve made of 100% recycled fiber. The shrinkwrap surrounding that package has also been replaced with a low-density LDPE polyethylene that's highly recyclable. Even the ink on the cardboard has been changed out for a vegetable-and-water-based version (so it's technically vegan if you're desperate for a snack).

The new packaging does cost a bit more to produce -- about 20 (British) cents per unit (or 30 percent), according to an open letter from Sports Interactive Studio Director Miles Jacobson. But those costs are somewhat offset by reduced shipping and destruction costs for excess units, he added. And as Spanish footballer Hector Bellerin says in a video accompanying the letter, "if there's no Earth, there's no money to spend." All told, Jacobson says the new packaging will save 55 grams of plastic per unit, or 20 tonnes across a print run of over 350,000. That's an extremely tiny dent in the estimated 335 million tons of plastic that is produced annually worldwide. But Jacobson hopes it could add up to a sizable dent if the entire industry follows suit for the tens of millions of discs it produces each year. "We're not the biggest game in the world," Jacobson said. "Imagine what happens if every other game, every film company, every music company switches to this packaging... So I'm throwing down the gauntlet here to ALL entertainment companies who use plastic for their Blu Ray, DVD and CD packaging."

Samsung has announced a new image sensor for phones that breaks records. Built-in partnership with Xiaomi, the new Samsung ISOCELL Bright HMX is the world's first mobile image sensor that goes beyond 100 million pixels. From a report: At 108MP, the new sensor allows for higher quality pictures in different light conditions. The resolution, which Samsung says is equivalent to DSLR cameras, allows for "extremely sharp photographs rich in detail," according to the firm. It's the first mobile image sensor to adopt a large lens size of 1/1.33-inch that allows the lens to absorb more light, leading to better quality pictures in low-light conditions. There's also an intelligent Tetracell technology that uses a pixel-merging method to "imitate" big-pixel sensors, allowing phones to produce brighter 27MP images. [...] The image sensor is built to tackle video recording as well, with Samsung claiming no losses in field-of-view when recording videos at resolutions up to 6K at 30fps.

The source of a gigantic, mysterious leak of radioactive material that swept across Europe in 2017 has been traced to a Russian nuclear facility, which appears to have been preparing materials for experiments in Italy. From a report: The leak released up to 100 times the amount of radiation into the atmosphere that the Fukushima disaster did. Italian scientists were the first to raise the alarm on 2 October, when they noticed a burst of the radioactive ruthenium-106 in the atmosphere. This was quickly corroborated by other monitoring laboratories across Europe. Georg Steinhauser at Leibniz University Hannover in Germany says he was "stunned" when he first noticed the event. Routine surveillance detects several radiation leaks each year, mostly of extremely low levels of radionuclides used in medicine. But this event was different. "The ruthenium-106 was one of a kind. We had never measured anything like this before," says Steinhauser. Even so, the radiation level wasn't high enough to impact human health in Europe, although exposure closer to the site of release would have been far greater.

Amazon workers around the world are going on strike today to bring attention to the working conditions they endure. "Some are arguing that buying from Amazon during Prime Day is akin to crossing a picket line," reports Quartz. From the report: As the two-day bacchanal of discounted Amazon offerings begins, workers at its fulfillment centers around the U.S. continue to complain of extremely odious quotas, limited bathroom breaks, mandatory holiday shifts, and the need for pain medication just to get through their 10-hour work days.

The U.S.: Workers at a Shakopee, Minnesota fulfillment center will be walking out during a six-hour period that overlaps with the end of the facility's morning shift and the start of its evening shift. There are about 1,500 full-time employees at the facility, according to the Daily Beast.

Germany: Hundreds of employees at seven facilities will be striking today and tomorrow, over longstanding issues with employee pay. âoeWhile Amazon holds a giant Prime-Day bargain hunt, employees are deprived of a living wage,â Orhan Akman, a representative from the German labor union Ver.di, said in a statement shared with Quartz.

The UK: The GMB trade union will be staging protests at Amazon facilities across the country. Some of the most shocking accounts issues of issues faced by Amazon warehouse workers have come out of the UK. One undercover writer said they witnessed co-workers urinating in bottles to avoid missing quotas by taking bathroom breaks.

Elsewhere in Europe: Workers in Spain and Poland will also be organizing demonstrations at Amazon facilities across their countries throughout the week. Here's what Amazon had to say about the demonstrations and walkouts: "Events like Prime Day have become an opportunity for our critics, including unions, to raise awareness for their cause, in this case, increased membership dues. These groups are conjuring misinformation to work in their favor, when in fact we already offer the things they purport to be their cause -- industry leading pay (full-time employees at our Shakopee facility make $16.25 - $20.80), benefits, and a safe workplace for our employees. We can only conclude that the people who plan to attend the event on Monday are simply not informed. If these groups -- unions and the politicians they rally to their cause -- really want to help the American worker, we encourage them to focus their energy on passing legislation for an increase in the federal minimum wage, because $7.25 is too low."

An anonymous reader quotes a report from NBC News: California regulators on Thursday approved allowing utilities to cut off electricity to possibly hundreds of thousands of customers to avoid catastrophic wildfires like the one sparked by power lines last year that killed 85 people and largely destroyed the city of Paradise. Utilities' liability can reach billions of dollars, and after several years of devastating wildfires, they asked regulators to allow them to pull the plug when fire risk is extremely high. That's mainly during periods of excessive winds and low humidity when vegetation is dried out and can easily ignite.

The California Public Utilities Commission gave the green light but said utilities must do a better job educating and notifying the public, particularly those with disabilities and others who are vulnerable, and ramp up preventive efforts, such as clearing brush and installing fire-resistant poles. The plans could inconvenience hundreds of thousands of customers while endangering some who depend on electricity to keep them alive. The precautionary outages could mean multiday blackouts for cities as large as San Francisco and San Jose, Northern California's major power provider warned in a recent filing with the utilities commission.

An anonymous reader quotes a report from Ars Technica: A report released today by the International Energy Agency (IEA) warns world leaders that -- without support for new nuclear power or lifetime extensions for existing nuclear power plants -- the world's climate goals are at risk. "The lack of further lifetime extensions of existing nuclear plants and new projects could result in an additional four billion tonnes of CO2 emissions," a press release from the IEA noted.

The report is the IEA's first report on nuclear power in two decades, and it paints a picture of low-carbon power being lost through attrition (due to the retirement of aging plants) or due to economics (extremely cheap natural gas as well as wind and solar undercutting more expensive nuclear power for years in some regions). Around the world, 452 nuclear reactors provided 2,700 terawatt-hours (TWh) of electricity in 2018. This makes nuclear a significant source of low-carbon energy on a global level. While 11.2 gigawatts (GW) of nuclear power were connected to the grid last year, all of the new capacity was located in China or Russia. "Without additional nuclear, the clean energy transition becomes more difficult and more expensive -- requiring $1.6 trillion of additional investment in advanced economies over the next two decades," IEA says. "Critically, a major clean energy shortfall would emerge by 2040, calling on wind and solar PV to accelerate deployment even further to fill the gap."

"When your phone rings, there's about a 50 percent chance it's a spam robo-call," reports the Washington Post. Now a computer science professor who's researched robo-call technologies reveals the economics behind automatically dialing phone numbers "either randomly, or from massive databases compiled from automated Web searches, leaked databases of personal information and marketing data." It doesn't matter whether you've signed up with the federal Do Not Call Registry, although companies that call numbers on the list are supposed to be subject to large fines. The robo-callers ignore the list, and evade penalties because they can mask the true origins of their calls.... Each call costs a fraction of a cent -- and a successful robo-call scam can net millions of dollars. That more than pays for all the calls people ignored or hung up on, and provides cash for the next round. Casting an enormous net at low cost lets these scammers find a few gullible victims who can fund the whole operation...

Partly that's because their costs are low. Most phone calls are made and connected via the Internet, so robo-call companies can make tens of thousands, or even millions, of calls very cheaply. Many of the illegal robo-calls targeting the United States probably come from overseas -- which used to be extremely expensive but now is far cheaper...

Meanwhile, the Federal Communications Commission has been asking U.S. phone companies to filter calls and police their own systems to keep out robo-calls. It hasn't worked, mainly because it's too costly and technically difficult for phone companies to do that. It's hard to detect fake Caller ID information, and wrongly blocking a legitimate call could cause them legal problems.
The professor's article suggests guarding your phone number like you guard your credit card numbers. "Don't give your phone number to strangers, businesses or websites unless it's absolutely necessary."

"Of course, your phone number may already be widely known and available, either from telephone directories or websites, or just because you've had it for many years. In that case, you probably can't stop getting robo-calls."

"Oumuamua, the first object ever seen passing through our solar system from interstellar space, was thought to be emitting gas like a comet to explain its weird motion," reports Syfy Wire, "but a new idea is that the comet is just very, very porous."

Astronomer Phil Plait writes: It was hard to tell what it was; it was too small, faint, and far away to get good observations, and worse, it was only seen on its way out, so it was farther from us literally every day. Then another very weird thing happened: More observations allowed a better determination of its trajectory, and it was found that it wasn't slowing down fast enough. As it moves away, the Sun's gravity pulls on it, slowing it down...but it wasn't slowing down enough. Some force was acting on it, accelerating it very slightly... A new paper has come out that might have a solution, and it's really clever. Maybe 'Oumuamua's not flat. Maybe it's fluffy... [And thus moved by the force of sunlight giving it a tiny push]

When stars are very young, they have a huge disk of material swirling around them; it's from this material that planets form. Out far from the star, where temperatures in the disk are cold, teeny tiny grains of dust and water ice can stick together in funny shapes, creating fractals... Materials made in a fractal pattern can be very porous, and in fact out in that protoplanetary disk around a young star, physical models show that objects can grow fractally until they're as big as 'Oumuamua, and have those extremely low densities needed to account for its weird behavior. So 'Oumuamua doesn't have to be a spaceship. It just has to be a snowflake! A three-dimensionally constructed phenomenally porous low-density snowflake... [T]he new paper suggests it came from a nearby star, and one that's relatively young (less than 100 million years). It formed out in the disk, and got ejected somehow, likely from a planet forming nearby giving it a boost from its gravity.
"I certainly hope we find more beasties like this one," Plait writes. "They can tell us so much about how planets form in other star systems, which is pretty hard to figure out from dozens or hundreds of light years away.

"It's a lot easier when they obligingly send bits of their building materials to us."

dryriver writes: A relative of mine has been left with extremely low speech volume -- about 25% of a normal speaker -- and lack of high pitch capability after a major throat surgery. He cannot speak on the phone at all now -- you cannot hear him properly on the other end of the line, even though you can understand his speech OK when you are standing in the room next to him. Is there an assistive Android app that can:
1. Significantly boost the output volume (e.g. X 4) of the Android phone microphone he speaks into.
2. Add voice box, equalizer, autotune or audio filtering/bass boosting type audio effects in realtime to the microphone input to fix the speaker's pitch.
3. Can filter out background noise to some extent (so it doesn't get boosted as well).
4. Allows these effects to be used easily during phonecalls?
All the Android microphone/equalizer/megaphone type apps I've tried so far have huge problems -- some are novelty voice-changer apps for teens, some demand ridiculously broad permissions to everything on the phone including realtime location data of the user, some demand that an external mic is attached to the phone, some are too simple technically to do anything useful and some are advertising-fests that are plain unusable. Is there a good Android -- or iOS -- app for the speech impaired that would give this person a chance to make audio phonecalls on a smartphone again? Thanks for any advice!

An anonymous reader writes: The best job for someone is not always in the area where they live. Often times, the job that will pay them most, and make the best use of their skills means moving to another city, state or country. Though making the choice to move can be difficult emotionally, it is extremely good for economic growth. Productive people make productive economies. Unfortunately for the US economy, people don't move they like they used to. According to recently released data from the US Census, only 10.1% of adults moved homes from August 2017 to August 2018. This is the lowest rate of moving since the government began collected data in 1948. The census tracks moves within counties, within states, or across states, and no matter how you look at it, moving rates are way down from just 15 years ago. For example, from 2002 to 2003, 2.8% of Americans moved across state lines. From 2017 to 2018, it was just 1.5%.

The value of Bitcoin has hit a new low of $4,951, bringing the total value of all Bitcoin in existence to below $87 billion. Much of the turmoil can be attributed to the split of Bitcoin Cash on November 15th. The Bitcoin offshoot has been split into two different cryptocurrencies, which are now in competition with each other. The BBC reports: Bitcoin exchange Kraken said in a blog post that it regarded one of the two new Bitcoin Cash crypto-currencies -- Bitcoin SV -- as "an extremely risky investment." At its peak, in November 2017, it briefly hit $19,783 - which means the price has fallen by about 75%. After the excitements of last year when the price soared to nearly $20,000 and then tumbled, Bitcoin has been rather dull and stable for much of 2018, settling between $6,000 and $7,000.

An anonymous reader shares a report: The ironic thing about the compressed state of air travel today is that planes are getting larger. The jet I was on, an Airbus A321, stretches nearly 23 feet longer than its predecessor, the A320. More space, more passengers, more profit. These bigger planes are increasingly the most common Âvariants -- both on American Airlines and across all carriers. The current Boeing 737s, the world's most flown craft, are all longer than the original by up to 45 feet. And yet, on the inside, we're getting squeezed.

That's because more space doesn't equal more space in Airline World. It equals more seats -- and typically less room per person. In 2017, for example, word leaked that American was planning to add six economy spots to its A320s, nine to its A321s, and 12 (that's two rows) to its Boeing 737-800s. JetBlue is reportedly ramming 12 extras into its A320s, and Delta's will gain 10. And, come 2020, you'll likely find more seats on every United plane. In Airline World, they call this densification, which is a silly word. Passengers call it arrrgh!

Consumer Reports recently polled 55,000 of its members about air travel. There were complaints about all aspects, from ticketing to agents checking carry-ons at the gate. But 30 percent of coach-class fliers rated their seats as outright uncomfortable, and every airline received extremely low scores on legroom and cushiness in economy. Clearly, things are dismal and seem to be getting even worse. They're so bad, in fact, that last year, nonprofit consumer-advocacy group FlyersRights.org filed a suit against the Federal Aviation Administration, after lobbying the agency to stop the squeeze and standardize seat sizes.

An anonymous reader quotes a report from IEEE Spectrum: Deep learning has a DRAM problem. Systems designed to do difficult things in real time, such as telling a cat from a kid in a car's backup camera video stream, are continuously shuttling the data that makes up the neural network's guts from memory to the processor. The problem, according to startup Flex Logix, isn't a lack of storage for that data; it's a lack of bandwidth between the processor and memory. Some systems need four or even eight DRAM chips to sling the 100s of gigabits to the processor, which adds a lot of space and consumes considerable power. Flex Logix says that the interconnect technology and tile-based architecture it developed for reconfigurable chips will lead to AI systems that need the bandwidth of only a single DRAM chip and consume one-tenth the power.

Mountain View-based Flex Logix had started to commercialize a new architecture for embedded field programmable gate arrays (eFPGAs). But after some exploration, one of the founders, Cheng C. Wang, realized the technology could speed neural networks. A neural network is made up of connections and "weights" that denote how strong those connections are. A good AI chip needs two things, explains the other founder Geoff Tate. One is a lot of circuits that do the critical "inferencing" computation, called multiply and accumulate. "But what's even harder is that you have to be very good at bringing in all these weights, so that the multipliers always have the data they need in order to do the math that's required. [Wang] realized that the technology that we have in the interconnect of our FPGA, he could adapt to make an architecture that was extremely good at loading weights rapidly and efficiently, giving high performance and low power."