Japan looks to create a superfast supercomputer for deep learning


Japan is reportedly eyeing a return to the top of the supercomputer ranks. The county’s Ministry of Economy, Trade and Industry plans to spend 19.5 billion yen ($173 million) on a new super computer, according to budget filings reported by Reuters, with aims of developing a machine capable of 130 petaflops.

That number would put the company in the top spot, moving the AI Bridging Cloud Infrastructure ahead of Sunway Taihulight, the supercomputer unveiled by China over the summer capable of 93 petaflops. The Ministry already has some fairly big plans for the previously unannounced super computer, utilizing its record-breaking speeds to help the country develop advances in AI technologies like deep learning.

Also on the list (at 130 quadrillion calculations per second, the thing would be fairly adept at multi-tasking) are gains in autonomous vehicle development, medicine and robotics. According to the report, the planned computer would be licensed to domestic corporations for a fee.

The Ministry’s not doing much talking about the computer at this early stage, but a director from the National Institute of Advanced Industrial Science and Technology, where the planned system would be constructed says, “as far as we know, there is nothing out there that is as fast.” Fujitsu’s Oakforest-PACS was certified as the country’s fastest computer earlier this month, capable of 13.6 petaflops.


Featured Image: Fujitsu’s previously announced PRIMEHPC FX10 supercomputer.

from TechCrunch http://ift.tt/2fzioPc

Mars probe crashed because it misjudged where the ground was


The European Space Agency believes that it knows what caused its Schiaparelli lander to crash on the surface of Mars. It turns out that the spacecraft was hurtling towards the ground perfectly well until it, uh, forgot where the ground actually was. A sensor tasked with determining its altitude failed for a single second, but that was long enough to wreck the entire mission. Since the vessel believed that it had already arrived, it ran through the rest of the landing process and activated the on-ground sensors.

The inquest as to what caused the failure is still ongoing and may not be resolved for the better part of a year. But officials believe that, specifically, Schiaparelli’s inertial measurement unit was oversaturated for a second, which told the system that it was already below ground. As a consequence, the lander fired its parachute, ejected its heat shield and fired its braking thrusters all at once while still 2.3 miles above the surface.

It’s the second time that Europe has attempted to land a vessel on Mars, only for something to go wrong at the last minute. Beagle 2, from 2003, which successfully arrived on the red planet, but failed to deploy its solar panels to power transmissions. In fact, the craft was only found in 2015, when a NASA spacecraft shot detailed images of the proposed landing site.

Europe’s space bods aren’t too worried about the failure of Schiaparelli, and will continue to work on the ExoMars program. It’s hoped that the next mission in the series will launch in 2020, hopefully with a computer that double-checks its math during a landing procedure.

Via: The Guardian

Source: ESA

from Engadget http://ift.tt/2fX9nQw

Hover Carts Make Plane Rides Easy



The most annoying part of any meal service mid-flight is when a passenger wants to get past the flight attendant and squeeze down the narrow aisle. The Cart line is a solution to this very common problem. You see, Murphy’s Law is always at play … you will always get the urge to use the restroom just when its next to impossible! So it’s either this functional cart that gets stowed overhead or upgrade to First Class.

Designers: Chiao-Chun Ni, SU-Min Wu, Kuan-Ting Chen & Yao-Ting Huang



from Yanko Design http://ift.tt/2fYY76u

DeepMind has hired AI safety experts to protect us from dangerous machines


Google DeepMind CEO Demis Hassabis

DeepMind has made a number of hires as part of an effort to mitigate the chance of its artificial intelligence developing into something dangerous, according to LinkedIn and other sources.

The London-based AI lab, which was acquired by Google in 2014 for £400 million, is building computer systems that can learn and think for themselves.

So far the company’s algorithms have been used to defeat humans at complex board games like Go and helped Google to cut its huge electricity bill. But DeepMind doesn’t plan to stop there; ultimately it wants to "solve intelligence" and use it to "make the world a better place."

In a bid to reduce the chance of creating dangerous artificial intelligence, DeepMind has hired Viktoriya Krakovna, Jan Leike, and Pedro Ortega into its AI safety group. It’s currently unclear when this group was formed.

Some of the world’s smartest minds, including physicist Stephen Hawking and Tesla founder Elon Musk, have warned that "superintelligent" machines — described in a book called "Superintelligence" by Oxford University philosopher Nick Bostrom — could end up being one of the greatest threats to humanity. They’re concerned that they could outsmart humans within a matter of decades and decide that we’re no longer necessary.

Krakovna, who joins DeepMind as a research scientist, holds a PhD in statistics from Harvard University and she cofounded the Future of Life Institute in Boston with MIT cosmologist Max Tegmark and Skype cofounder Jaan Tallinn.

The institute, which counts Hawking and Musk as board advisors, was created to mitigate existential risks facing humanity, particularly existential risk from advanced artificial intelligence (AI).

While at DeepMind, the former Google engineer will carry out technical research on AI safety, according to her LinkedIn profile.

Leike has also joined DeepMind as a research scientist, according to his personal website.

In addition to his role at DeepMind Leike is also a research associate at Oxford University’s Future of Humanity Institute, which is lead by Bostrom.

On his website, Leike writes: "My research aims at making machine learning robust and beneficial. I work on problems in reinforcement learning orthogonal to capability: How do we design or learn a good objective function? How can we design agents such that they are incentivised to act in our best interests? How can we avoid degenerate solutions to the objective function?"

Ortega, also a research scientist at DeepMind, holds a PhD in machine learning from Cambridge University. According to a short bio on his personal website: "His work includes the application of information-theoretic and statistical mechanical ideas to sequential decision-making."

DeepMind did not immediately respond to Business Insider’s request for comment.

Join the conversation about this story »

NOW WATCH: The ‘Apple of China’ just unveiled a phone that’s more powerful and better looking than the iPhone

from SAI http://ift.tt/2fvwJco

NASA says the ISS’ inflatable module is doing great


It’s been almost six months since the Bigelow Expandable Activity Module (BEAM) was attached to the ISS, and NASA now has some early data to share. According to BEAM Manager Steve Munday, it’s been doing well and performing as expected after the installation hiccup that prevented it from unfolding the first time. To start with, NASA Langley scientists didn’t find any sign of large debris impact that could affect the module’s ability to protect inhabitants. The folks over at NASA’s Johnson Space Center didn’t detect abnormally high radiation levels inside the habitat, as well — in fact, it exhibited levels similar to the rest of the space station’s.

In addition, it was warmer than expected (though still quite cold) inside the module, probably because its layers provide more insulation than Bigelow Aerospace scientists thought they would. That’s not an issue, though, since its designers were actually aiming for a warm interior. "A colder-than-expected BEAM would have increased the risk of condensation," Munday explained, "so we were pleased when Jeff first opened the hatch and found the interior to be bone dry."

Munday was talking about Jeff Williams, one of the most experienced NASA astronauts and one of the first two people who stepped foot inside the module. See, the plan is for ISS crew members to enter the habitat and check on its condition 67 or so times a year within its two-year test period. Williams and Russian cosmonaut Oleg Skripochka collected air sample and downloaded data from all the sensors inside back in June.

While the results sound promising, it’s too early to tell whether Bigelow Aerospace’s dream of deploying large inflatable space stations to Low Earth Orbit is feasible. If you’ll recall, the private space corp teamed up with United Launch Alliance to develop expandable modules larger than BEAM that can orbit the Earth on their own. If the partners succeed, the stations will be much cheaper to launch than similarly sized rigid structures, since they can be folded inside their carriers. As Munday said, BEAM is the first of its kind. The scientists involved are still learning, and they’ll no doubt incorporate changes to the technology as they learn more about the expandable habitat.

Source: NASA

from Engadget http://ift.tt/2gGlVNo