Touching the Future | Ray Kurzweil | SU Global Summit

Advertisements

http://appleinsider.com/articles/17/10/31/apple-ios-111-now-available-for-iphone-ipad-ipod-touch-with-70-new-emoji-other-improvements

Apple iOS 11.1 now available for iPhone, iPad, iPod touch with 70 new emoji, other improvements

After a month-long testing process, Apple has made available iOS 11.1 including a fix for the KRACK wi-fi exploit, and a large amount of accessibility feature fixes.

In addition to the KRACK wi-fi attack vector rectification, Apple’s iOS 11.1 also includes the return of the 3D Touch multitasking app switcher.

Apple’s notes on the update proclaim that the update “introduces over 70 new emoji.” The new emoji require iOS 11.1 devices for the intended recipient as well, or the graphic in question won’t be properly displayed.

Other fixes include resolution of several issues in Photos, improved Accessibility with many VoiceOver enhancements and improved braille device support, some Apple Watch data handoff issues repairs, and a problem resulting in inaccuracies in location data from external GPS devices has been patched.

Still absent are Apple Pay Cash, and AirPlay 2.

Three “sub-point” updates proceeded the iOS 11.1 release, fixing some glaring bugs —all of which have been rolled into iOS 11.1.

The update is a 311.3 MB download on an iPhone 7 Plus.

http://linuxgizmos.com/rpi-zero-w-clone-offers-quad-core-power-for-15/

Raspberry Pi Zero W clone offers quad-core power for $15

SinoVoip’s Linux-friendly, 60 x 30mm Banana Pi M2 Zero (BPI-M2 Zero) SBC closely mimics the Raspberry Pi Zero W, but has a faster Allwinner H2+.

Just as we were trumpeting the $23 BPI-M2 Magic as being the “smallest, cheapest Banana Pi yet,” SinoVoip has launched an even tinier and more affordable Linux/Android hacker board on AliExpress. The WiFi-enabled Banana Pi M2 Zero (BPI-M2 Zero), which was revealed back in July, is now selling for only $15 with the standard 512MB RAM, or $21.53 including shipping to the U.S.

 
BPI-M2 Zero, front and back
(click images to enlarge)
The BPI-M2 Zero SBC openly mimics the $10 Raspberry Pi Zero W in both dimensions and features, but offers a 1.2GHz, quad-core, Cortex-A7 Allwinner H2+ compared to the RPi Zero W’s single-core, 1GHz ARM11-based Broadcom BCM2836. Like other Banana Pi boards, it is an open source hardware and software design with community support.The Allwinner H2+ is much like the widely adopted Allwinner H3 SoC except that it tops out at HD resolution instead of 4K. This is the same SoC found on two other RPi Zero W clones: FriendlyElec’s $8 (256MB) to $12 (512MB) NanoPi Duo and Shenzhen Xunlong’s $7 to $12.30 (512MB) Orange Pi Zero.


BPI-M2 Zero (bottom) with Raspberry Pi Zero W (top)
(click image to enlarge)
The BPI-M2 Zero measures 60 x 30m (1,800 square mm) compared to 65 x 30mm (1,950 sq. millimeters) for the RPi Zero W. Since the layout is almost identical, you can use a Zero W case.By comparison, the NanoPi Duo measures 50 x 25.4mm (1,270 sq. mm) and the Orange Pi Zero is 48 x 46mm (2,208 sq. mm). The recently released BPI-M2 Magic has a footprint of 51 x 51mm (2,500 sq. mm). The BPI-M2 Zero’s 35-gram weight, however, is considerably greater than its rivals.

The BPI-M2 Zero is a much closer clone of the Raspbery Pi Zero W than the other imitators. It has a standard allotment of 512MB RAM rather than a 256MB entry point like the NanoPi Duo and Orange Pi Zero, and unlike those two boards, it offers Bluetooth in addition to WiFi.


BPI-M2 Zero detail view
(click image to enlarge)
The SBC is further equipped with a mini-HDMI port, microSD slot, micro-USB OTG port, and 5V power-only micro-USB port. The BPI-Zero lacks the composite video header of the Zero W, but similarly provides a CSI camera connector. Also like the Zero W, it has a 40-pin GPIO header compared to the 32- and 24-pin headers for the NanoPi Duo and Orange Pi Zero, respectively.Specifications are listed for the BPI-M2 Zero include:

  • Processor — Allwinner H2+ (4x Cortex-A7 @ 1.2GHz); ARM Mali-400 MP2 GPU @600MHz
  • Memory — 512MB DDR3 SDRAM
  • Storage — MicroSD slot for up to 64GB
  • Wireless:
    • 802.11b/g/n plus Bluetooth 4.0 dual mode (Ampak AP6212)
    • Optional dual mode Broadcom AP6335 with 802.11ac and BT 4.0 or Gigafu Tech AP6181 (2.4GHz WiFi and no BT, but low power consumption)
    • RF connector
  • Display — Mini-HDMI port with audio for up to 1080p60
  • Other I/O:
    • Micro-USB 2.0 OTG port (with power support)
    • MIPI-CSI for 5MP cam or 1080p @30 video input
    • Debug UART/ground header with 3x GPIO
    • 40-pin RPi 3-compatible expansion connector
  • Other features — 2x LEDs; power and reset buttons
  • Power — 5V/2A via micro-USB
  • Dimensions — 60 x 30mm
  • Weight — 35 g
  • Operating system — Android, Debian, Ubuntu, Raspbian image

Further information

The Banana Pi M2 Zero (BPI-M2 Zero) is available for $15, or $21.53 including shipping to the U.S. More information may be found on the BPI-M2 Zero AliExpress shopping page and Banana Pi wiki.

http://www.imeche.org/news/news-article/spider-silk-could-improve-microphones-and-hearing-aids1

Spider silk could improve microphones and hearing aids

Amit Katwala

(Credit: iStock)
(Credit: iStock)

Fine fibres such as spider silk could lead to new and better microphones that sense airflow fluctuations rather than pressure changes.

Some insects, including mosquitos, flies and spiders, sense sound using fine hairs on their bodies that move with the sound waves travelling through the air. “We use our eardrums which pick up the direction of sound based on pressure, but most insects actually hear with their hairs,” explained Ron Miles, a professor at Binghamton University in New York.

Working alongside graduate student Jian Zhou, Miles recreated a similar system inside a microphone, which had better directional sensing across a wider range of frequencies than traditional models. It could give hearing aid or smartphone users the ability to cancel out background noise more effectively when having a conversation in a crowed area.

To create their microphone, Miles and Zhou used spider silk, which is thin enough that it moves with the air when hit by soundwaves. “This can even happen with infrasound at frequencies as low as 3 Hertz,” said Miles – that’s the equivalent of hearing the normally inaudible rumble of tectonic plates moving in an earthquake.

To translate the movement of the spider silk into an electronic signal, the researchers coated it with gold and placed it in a magnetic field. “It’s actually a fairly simple way to make an extremely effective microphone that microphone that has better directional capabilities across a wide range of frequencies,” said Miles.

Rob Malkin, an expert in bio-inspired acoustic devices from the University of Bristol, told Professional Engineering that the research demonstrated yet again “how a beautiful design from the insect world can lead to advancements in microphone engineering”.

He called the work a “step change” in how microphones could function in the future, as it expanded the narrow range of frequencies that insects can hear at, into a spectrum broad enough for humans. “The work is very encouraging as it shows that the physical process being exploited by many insects – that is hearing with hairs – is relatively simple, and should make the manufacture of broadband devices finally possible,” Malkin added.

http://www.kurzweilai.net/a-tool-to-debug-black-box-deep-learning-neural-networks

A tool to debug ‘black box’ deep-learning neural networks

Brings transparency to self-driving cars and other self-taught systems
October 30, 2017

Oops! A new debugging tool called DeepXplore generates real-world test images meant to expose logic errors in deep neural networks. The darkened photo at right tricked one set of neurons into telling the car to turn into the guardrail. After catching the mistake, the tool retrains the network to fix the bug. (credit: Columbia Engineering)

Researchers at Columbia and Lehigh universities have developed a method for error-checking the reasoning of the thousands or millions of neurons in unsupervised (self-taught) deep-learning neural networks, such as those used in self-driving cars.

Their tool, DeepXplore, feeds confusing, real-world inputs into the network to expose rare instances of flawed reasoning, such as the incident last year when Tesla’s autonomous car collided with a truck it mistook for a cloud, killing its passenger. Deep learning systems don’t explain how they make their decisions, which makes them hard to trust.

Modeled after the human brain, deep learning uses layers of artificial neurons that process and consolidate information. This results in a set of rules to solve complex problems, from recognizing friends’ faces online to translating email written in Chinese. The technology has achieved impressive feats of intelligence, but as more tasks become automated this way, concerns about safety, security, and ethics are growing.

Finding bugs by generating test images

Debugging the neural networks in self-driving cars is an especially slow and tedious process, with no way to measure how thoroughly logic within the network has been checked for errors. Current limited approaches include randomly feeding manually generated test images into the network until one triggers a wrong decision (telling the car to veer into the guardrail, for example); and “adversarial testing,” which automatically generates test images that it alters incrementally until one image tricks the system.

The new DeepXplore solution — presented Oct. 29, 2017 in an open-access paper at ACM’s Symposium on OperatingSystems Principles in Shanghai — can find a wider variety of bugs than random or adversarial testing by using the network itself to generate test images likely to cause neuron clusters to make conflicting decisions, according to the researchers.

To simulate real-world conditions, photos are lightened and darkened, and made to mimic the effect of dust on a camera lens, or a person or object blocking the camera’s view. A photo of the road may be darkened just enough, for example, to cause one set of neurons to tell the car to turn left, and two other sets of neurons to tell it to go right.

After inferring that the first set misclassified the photo, DeepXplore automatically retrains the network to recognize the darker image and fix the bug. Using optimization techniques, researchers have designed DeepXplore to trigger as many conflicting decisions with its test images as it can while maximizing the number of neurons activated.

“You can think of our testing process as reverse-engineering the learning process to understand its logic,” said co-developer Suman Jana, a computer scientist at Columbia Engineering and a member of the Data Science Institute. “This gives you some visibility into what the system is doing and where it’s going wrong.”

Testing their software on 15 state-of-the-art neural networks, including Nvidia’s Dave 2 network for self-driving cars, the researchers uncovered thousands of bugs missed by previous techniques. They report activating up to 100 percent of network neurons — 30 percent more on average than either random or adversarial testing — and bringing overall accuracy up to 99 percent in some networks, a 3 percent improvement on average.*

The ultimate goal: certifying a neural network is bug-free

Still, a high level of assurance is needed before regulators and the public are ready to embrace robot cars and other safety-critical technology like autonomous air-traffic control systems. One limitation of DeepXplore is that it can’t certify that a neural network is bug-free. That requires isolating and testing the exact rules the network has learned.

A new tool developed at Stanford University, called ReluPlex, uses the power of mathematical proofs to do this for small networks. Costly in computing time, but offering strong guarantees, this small-scale verification technique complements DeepXplore’s full-scale testing approach, said ReluPlex co-developer Clark Barrett, a computer scientist at Stanford.

“Testing techniques use efficient and clever heuristics to find problems in a system, and it seems that the techniques in this paper are particularly good,” he said. “However, a testing technique can never guarantee that all the bugs have been found, or similarly, if it can’t find any bugs, that there are, in fact, no bugs.”

DeepXplore has applications beyond self-driving cars. It can find malware disguised as benign code in anti-virus software, and uncover discriminatory assumptions baked into predictive policing and criminal sentencing software, for example.

The team has made their open-source software public for other researchers to use, and launched a website to let people upload their own data to see how the testing process works.

* The team evaluated DeepXplore on real-world datasets including Udacity self-driving car challenge data, image data from ImageNet and MNIST, Android malware data from Drebin, PDF malware data from Contagio/VirusTotal, and production-quality deep neural networks trained on these datasets, such as these ranked top in Udacity self-driving car challenge. Their results show that DeepXplore found thousands of incorrect corner case behaviors (e.g., self-driving cars crashing into guard rails) in 15 state-of-the-art deep learning models with a total of 132,057 neurons trained on five popular datasets containing around 162 GB of data.


Abstract of DeepXplore: Automated Whitebox Testing of Deep Learning Systems

Deep learning (DL) systems are increasingly deployed in safety- and security-critical domains including self-driving cars and malware detection, where the correctness and predictability of a system’s behavior for corner case inputs are of great importance. Existing DL testing depends heavily on manually labeled data and therefore often fails to expose erroneous behaviors for rare inputs.

We design, implement, and evaluate DeepXplore, the first whitebox framework for systematically testing real-world DL systems. First, we introduce neuron coverage for systematically measuring the parts of a DL system exercised by test inputs. Next, we leverage multiple DL systems with similar functionality as cross-referencing oracles to avoid manual checking. Finally, we demonstrate how finding inputs for DL systems that both trigger many differential behaviors and achieve high neuron coverage can be represented as a joint optimization problem and solved efficiently using gradient-based search techniques.

DeepXplore efficiently finds thousands of incorrect corner case behaviors (e.g., self-driving cars crashing into guard rails and malware masquerading as benign software) in state-of-the-art DL models with thousands of neurons trained on five popular datasets including ImageNet and Udacity self-driving challenge data. For all tested DL models, on average, DeepXplore generated one test input demonstrating incorrect behavior within one second while running only on a commodity laptop. We further show that the test inputs generated by DeepXplore can also be used to retrain the corresponding DL model to improve the model’s accuracy by up to 3%.

http://uk.businessinsider.com/apple-homepod-spotify-2017-10

Spotify won’t work with Siri on Apple’s HomePod

Apple HomePod white and blackAP

  • Apple’s smart speaker, HomePod, will go on sale in December for $349. People will use it by talking to it. 
  • Apple wants developers to make Siri apps for the HomePod, but only in a few categories — messaging, lists, and notes.
  • This means that users won’t be able to tell the HomePod to play music from Spotify, although Apple’s streaming service, Apple Music, is supported. 

 

Apple encouraged developers on Monday to make apps for HomePod, Apple’s new smart speaker that will go on sale before the end of the year.

Because the HomePod does not have a screen, users are expected to interact with it primarily by talking to it and to Apple’s voice assistant, Siri.

But the third-party apps that Apple is opening the doors for on the HomePod are fairly limited — Apple says that only apps that revolve around messaging, lists, and notes can integrate with Siri on the HomePod and will use a nearby iPhone or iPad to process commands.

The lack of music app support means that users won’t be able to play Spotify on the HomePod the way it was intended — by using your voice to tell it to play a song. (Users also won’t be able to order an Uber or Lyft by speaking to the HomePod, or make a call on Skype, based on the limited app categories.)

Apple Music users, of course, can call up their favorite tracks or albums by speaking to the smart speaker.

“We are always working to have Spotify available across all platforms, but we don’t have any further information to share at this time,” a Spotify representative told Business Insider, pointing out that users will be able to play Spotify on the HomePod speaker by using the AirPlay feature from an iPhone or iPad. Users will control the music by tapping around the Spotify smartphone app, rather than by using verbal commands.

“Third party apps like Spotify would just play music on HomePod using AirPlay 2, as we said back in June,” an Apple representative told Business Insider. AirPlay enables HomePod to be used like other wireless speakers.

When Apple first opened up Siri to third-party developers it took a similar approach. At first, Apple limited Siri access to apps only in a few categories, such as ride-hailing and fitness, although it expanded the number of categories earlier this summer.

However, Siri still does not support music apps on the iPhone and Spotify has no Siri features.

http://wccftech.com/best-premium-iphone-ipad-apps-gone-free-today-download-them-all/

Best Premium iPhone & iPad Apps Gone FREE Today – Download them All

These are the best premium iPhone and iPad apps that have gone free for a limited period of time. Grab them from here before they return to their original prices.

Paid iPhone & iPad Apps Have Gone FREE. Download them All Today.

Today is a great day if you are asking us. Though it’s a Monday, but some great premium iPhone and iPad apps have gone absolutely free for everyone to download. So, the deal is simple, just pick up your phone or tablet and start downloading these wonderful offers before they return to their original prices.