Google DeepMind’s WaveNet gets closer to replicating human speech

DeepMind Technologies logo

Google has been known to be fond of artificial intelligence, thanks to their various attempts to improve such technologies. The Search Giant has been directly involved with chatbots. They are also involved with AI robots that can write eerie post-modern poetry. And the company knows your fear of a robot apocalypse, so they have also created a system that would stop these machines from causing any harm.

But despite all these improvements, AI assistants are far from sounding like real humans. AI speech sounds very robotic, something that is vastly being improved by WaveNet, Deepmind’s new AI that can mimic human speech. It is not perfect, but is 50% better than current technologies. In fact, the thing is so smart it can create its own music, after learning various classical piano songs.

You can listen to some samples in DeepMind’s blog post. They are really quite impressive, but you likely won’t be seeing this hit the market soon, mostly because it requires too much computing power.

Researchers usually avoid modelling raw audio because it ticks so quickly: typically 16,000 samples per second or more, with important structure at many time-scales. Building a completely autoregressive model, in which the prediction for every one of those samples is influenced by all previous ones (in statistics-speak, each predictive distribution is conditioned on all previous observations), is clearly a challenging task.

For those out of the loop, Deepmind was acquired by Google in 2014 for $500 million. The Google-owned company’s system tries to mimic how the human mind works. It can be trained to learn information and has been known to beat Go champions, a great accomplishment considering this has been long known to be a distinctly human game.

Only time will tell how this will all pan out, but for now we can keep our eyes open to see how it all unfolds. Maybe soon enough you will be able to have actual conversations with your devices!

Internet Users in India May Reach 500 Million in 5-8 Months

Internet Users in India May Reach 500 Million in 5-8 Months: Prasad

The number of Internet users in the country is expected to go up to 500 million, from the present over 400 million, in next 5-8 months, Telecom Minister Ravi Shankar Prasad said on Sunday.

“The number of people using Internet has crossed 400 million in the country at present. I am expecting this number to go up to 500 million in next 5-8 months… According to our estimate, out of these, 70 percent people are going to access Internet through mobile phones,” Prasad said in Indore at an Indore Management Association function.

About 1 billion people, out of the population of India of nearly 1.25 billion, have mobile phones, the minister said, adding that every month, 2-3.5 million mobile users are added.

“The NDA government, led by Prime Minister Narendra Modi, is trying to awaken the efficiency and innovation in the public. We are constantly simplifying the government policies and procedures for good governance and development,” he said.

“Former Prime Minister Atal Bihari Vajpayee’s government is known for construction of National Highways. The present Prime Minister Narendra Modi government will be known for setting up Information Highways,” Prasad said.

The Telecom and Information Technology Minister also said that there are about 950 million Aadhar card holders in the country and the government has made the Aadhar card mandatory for reaping the benefits of many schemes to promote good governance.

If the Supreme Court gives nod, then the Aadhar card will be made mandatory for all government schemes, he added.

Prasad also said that the government is working towards connecting 250,000 gram panchayats in the country through optical fibre network.

When all the gram panchayats would be linked through this network, then e-business, e-education, e-health, and other projects could be started in villages, he added.

How the brain can handle so much data

Researchers at Georgia Tech discovered that humans can categorize data using less than 1 percent of the original information, and validated an algorithm to explain human learning — a method that also can be used for machine learning, data analysis and computer vision.

“How do we make sense of so much data around us, of so many different types, so quickly and robustly?” said Santosh Vempala, Distinguished Professor of Computer Science at the Georgia Institute of Technology and one of four researchers on the project. “At a fundamental level, how do humans begin to do that? It’s a computational problem.”

Researchers Rosa Arriaga, Maya Cakmak, David Rutter, and Vempala at Georgia Tech’s College of Computing studied human performance in “random projection” tests to understand how well humans learn an object. They presented test subjects with original, abstract images and then asked whether they could correctly identify that same image when randomly shown just a small portion of it.

“We hypothesized that random projection could be one way humans learn,” Arriaga, a senior research scientist and developmental psychologist, explains. “The short story is, the prediction was right. Just 0.15 percent of the total data is enough for humans.”

Next, researchers tested a computational algorithm to allow machines (very simple neural networks) to complete the same tests. Machines performed as well as humans, which provides a new understanding of how humans learn. “We found evidence that, in fact, the human and the neural network behave very similarly,” Arriaga said.

The researchers wanted to come up with a mathematical definition of what typical and atypical stimuli look like and, from that, predict which data would hardest for the human and the machine to learn. Humans and machines performed equally, demonstrating that indeed one can predict which data will be hardest to learn over time.

Results were recently published in the journal Neural Computation (MIT press). It is believed to be the first study of “random projection,” the core component of the researchers’ theory, with human subjects.

To test their theory, researchers created three families of abstract images at 150 x 150 pixels, then very small “random sketches” of those images. Test subjects were shown the whole image for 10 seconds, then randomly shown 16 sketches of each. Using abstract images ensured that neither humans nor machines had any prior knowledge of what the objects were.

“We were surprised by how close the performance was between extremely simple neural networks and humans,” Vempala said. “The design of neural networks was inspired by how we think humans learn, but it’s a weak inspiration. To find that it matches human performance is quite a surprise.”

“This fascinating paper introduces a localized random projection that compresses images while still making it possible for humans and machines to distinguish broad categories,” said Sanjoy Dasgupta, professor of computer science and engineering at the University of California San Diego and an expert on machine learning and random projection. “It is a creative combination of insights from geometry, neural computation, and machine learning.”

Although researchers cannot definitively claim that the human brain actually engages in random projection, the results support the notion that random projection is a plausible explanation, the authors conclude. In addition, it suggests a very useful technique for machine learning: large data is a formidable challenge today, and random projection is one way to make data manageable without losing essential content, at least for basic tasks such as categorization and decision making.

The algorithmic theory of learning based on random projection already has been cited more than 300 times and has become a commonly used technique in machine learning to handle large data of diverse types.

Scientists teach machines to learn like humans

“Our results show that by reverse engineering how people think about a problem, we can develop better algorithms,” explains Brenden Lake, a Moore-Sloan Data Science Fellow at New York University and the paper’s lead author. “Moreover, this work points to promising methods to narrow the gap for other machine learning tasks.”

The paper’s other authors were Ruslan Salakhutdinov, an assistant professor of Computer Science at the University of Toronto, and Joshua Tenenbaum, a professor at MIT in the Department of Brain and Cognitive Sciences and the Center for Brains, Minds and Machines.

When humans are exposed to a new concept — such as new piece of kitchen equipment, a new dance move, or a new letter in an unfamiliar alphabet — they often need only a few examples to understand its make-up and recognize new instances. While machines can now replicate some pattern-recognition tasks previously done only by humans — ATMs reading the numbers written on a check, for instance — machines typically need to be given hundreds or thousands of examples to perform with similar accuracy.

“It has been very difficult to build machines that require as little data as humans when learning a new concept,” observes Salakhutdinov. “Replicating these abilities is an exciting area of research connecting machine learning, statistics, computer vision, and cognitive science.”

Salakhutdinov helped to launch recent interest in learning with ‘deep neural networks,’ in a paper published in Science almost 10 years ago with his doctoral advisor Geoffrey Hinton. Their algorithm learned the structure of 10 handwritten character concepts — the digits 0-9 — from 6,000 examples each, or a total of 60,000 training examples.

In the work appearing in Science this week, the researchers sought to shorten the learning process and make it more akin to the way humans acquire and apply new knowledge — i.e., learning from a small number of examples and performing a range of tasks, such as generating new examples of a concept or generating whole new concepts.

To do so, they developed a ‘Bayesian Program Learning’ (BPL) framework, where concepts are represented as simple computer programs. For instance, the letter ‘A’ is represented by computer code — resembling the work of a computer programmer — that generates examples of that letter when the code is run. Yet no programmer is required during the learning process: the algorithm programs itself by constructing code to produce the letter it sees. Also, unlike standard computer programs that produce the same output every time they run, these probabilistic programs produce different outputs at each execution. This allows them to capture the way instances of a concept vary, such as the differences between how two people draw the letter ‘A.’

While standard pattern recognition algorithms represent concepts as configurations of pixels or collections of features, the BPL approach learns “generative models” of processes in the world, making learning a matter of ‘model building’ or ‘explaining’ the data provided to the algorithm. In the case of writing and recognizing letters, BPL is designed to capture both the causal and compositional properties of real-world processes, allowing the algorithm to use data more efficiently. The model also “learns to learn” by using knowledge from previous concepts to speed learning on new concepts — e.g., using knowledge of the Latin alphabet to learn letters in the Greek alphabet. The authors applied their model to over 1,600 types of handwritten characters in 50 of the world’s writing systems, including Sanskrit, Tibetan, Gujarati, Glagolitic — and even invented characters such as those from the television series Futurama.

In addition to testing the algorithm’s ability to recognize new instances of a concept, the authors asked both humans and computers to reproduce a series of handwritten characters after being shown a single example of each character, or in some cases, to create new characters in the style of those it had been shown. The scientists then compared the outputs from both humans and machines through ‘visual Turing tests.’ Here, human judges were given paired examples of both the human and machine output, along with the original prompt, and asked to identify which of the symbols were produced by the computer.

While judges’ correct responses varied across characters, for each visual Turing test, fewer than 25 percent of judges performed significantly better than chance in assessing whether a machine or a human produced a given set of symbols.

“Before they get to kindergarten, children learn to recognize new concepts from just a single example, and can even imagine new examples they haven’t seen,” notes Tenenbaum. “I’ve wanted to build models of these remarkable abilities since my own doctoral work in the late nineties. We are still far from building machines as smart as a human child, but this is the first time we have had a machine able to learn and use a large class of real-world concepts — even simple visual concepts such as handwritten characters — in ways that are hard to tell apart from humans.”

The work was supported by grants from the National Science Foundation to MIT’s Center for Brains, Minds and Machines (CCF-1231216), the Army Research Office (W911NF-08-1-0242, W911NF-13-1-2012), the Office of Naval Research (N000141310333), and the Moore-Sloan Data Science Environment at New York University.

Ronda Rousey, Taylor Swift, and Syrian refugees dominate Bing’s top searches of 2015

bing2015

In just a little over four weeks we’ll welcome in 2016 and bid adieu to the year that was. That means it’s time for Microsoft and Google to share break out the top searches of the year and show what caught the world’s attention in 2015.

First: Microsoft’s top Bing searches of 2015. Microsoft divvied up its search lists by celebrities, sports, and news to provide a more granular look at the information we collectively sought out in 2015.

Here are some of the highlights.

Sporting women

Women’s sports gripped the world in 2015, according to Bing. During the summer, the U.S. women’s national soccer team went on a tear to win the FIFA Women’s World Cup with a 5-2 victory over Japan in the final.

Serena Williams came close to taking all four Grand Slam tournaments in 2015, but she ultimately fell to Roberta Vinci at the U.S. Open in the semifinal. Thirty-four year-old Williams’ drive attracted so much interest that searches for her were greater than the top five men’s players searches combined.

Lastly, the UFC mixed martial arts match-up between Ronda Rousey and Holly Holm was the second most searched-for fight overall in 2015 after the so-called fight of the century between Floyd Mayweather and Manny Pacquiao. Holm scored an upset knockout win over a seemingly unbeatable Rousey in November in Australia.

Top News searches

This year’s tragedies in Paris and the ongoing suffering perpetrated by the Islamic State topped Bing’s news searches for the year. That was closely followed by the EU refugee crisis, which really grabbed the world’s attention in September after the body of three year-old Alan Kurdi was found on a beach in Turkey.

Following behind Isis and Syria was the 7.8 magnitude earthquake in Nepal on April 25 that killed more than 9,000 people and displacing more than 450,000. The major snowstorms that covered the Northeastern United States in early 2015, and the Germanwings 9525 helicopter crash in the French Alps rounded out the top five news searches of the year in the U.S.

The Celebs

Any year-in-search review would not be complete without a quick look at the celebrities we obsessed over in 2015. Top of the list was Caitlyn Jenner, who broke boundaries when she became the first transgender woman to appear on the cover of Vanity Fair. Jenner also sparked a public conversation about gender identity.

Following Jenner, was Miley Cyrus, Taylor Swift, Kim Richards, and Kim Kardashian. Swift got a lot of people talking in tech circles this year after she followed up on her 2014 Spotify dis by getting Apple to shape up with its royalties to artists during the Apple Music free trial period. Plus, who doesn’t love the Swift on Security parody Twitter account?

There’s a lot more to Bing’s look back at 2015 including the top sports moments and Bing’s “strange but true” stories. You can find it all on Bing Trends.

Yahoo changes direction, spins off everything except Alibaba

151209 yahoo cnbc

Yahoo announced today that it has abandoned a plan to spin off its $31 billion stake in Alibaba, a popular Chinese e-commerce company, and would instead spin off all of its other assets into a new company. The move is called a reverse spinoff.

The company said in a statement that concerns over a massive tax liability connected to an Alibaba spinoff forced them to nix that idea.

“Informed by our intimate familiarity with Yahoo’s unique circumstances, the board remains committed to accomplishing the significant business purposes and shareholder benefits that can be realized by separating the Alibaba stake from the rest of Yahoo,” said Maynard Webb, chairman of Yahoo’s board of directors. “To achieve this, we will now focus our efforts on the reverse spin off plan.”

Webb emphasized in an interview on CNBC this morning that Yahoo’s plan to spin off its core Internet business into a separate company does not mean it is actively trying to sell that business.

However, if a good offer comes along, the company would consider it. “The board has a fiduciary obligation to look at any good offer,” Webb said. “We are not proactively trying to do any of that.”

After weeks of speculation that Yahoo CEO Marissa Mayer might be forced out after not making enough headway in turning the company around in her three and a half years at the helm, it appears that her position is solid … at least for now.

The board did not make an announcement connected to Mayer and she specifically mentioned her alignment with the board in this morning’s conference call.

“Today I’m leading a very different company than the one I started at,” Mayer said, adding that the company today has better products, more modern advertising offerings and more focused employees. “I remain aligned with our board and our management team. We’re focused on shipping great products and features. I remain convinced that Yahoo is on a better path and the right one.”

Webb also took a direct question on CNBC about Mayer staying with the company.

“Absolutely,” he said. “I’ve never met anybody who works harder, is smarter and cares more. We want to help her return this company to an iconic place where it belongs. We have hundreds of millions of consumers that use our products every day, and we strive to make those products better and we need to find a way to capture that value and increase it. And I’m convinced we can.”

Mayer also said she’s “happy with the achievements” that the company has made under her leadership. She specifically pointed to moves she has made in mobile, video and social media.

“In addition to our efforts to increase value and diminish uncertainty for investors, the ultimate separation of our Alibaba stake will be important to our continued business transformation,” Mayer said. “In 2016, we will tighten our focus and prioritize investments to drive profitability and long-term growth. A separation from our Alibaba stake, via the reverse spin, will provide more transparency into the value of Yahoo’s business.”

Yahoo’s Chief Financial Officer Ken Goldman, also on the call this morning, said that spinning off Yahoo’s core Internet business is expected to require third-party consents, shareholder approval and filings and clearance from the Securities and Exchange Commission.

He also said the reverse spinoff could take a year or more to complete.

Google Photos keeps it simple with shared albums

googlephotossharedalbum

Google Photos is trying to streamline group photo albums by allowing anyone with a link to contribute their own pictures.

With the latest Google Photos update for iOS, Android, and the web, users can create a shared album from any cluster of photos. This generates a link, which can be copied, pasted, or shared to other apps and social networks. Once someone else has that link, they can upload pictures and add the collection to their own Google Photos library.

Compared to other image sharing platforms, Google’s is decidedly bare-bones. There’s no way to comment or “like” an image, as you can with Facebook or iCloud Photo Sharing, and there are no options for face tagging beyond the automatic (and private) image recognition that Google Photos performs automatically.

But maybe that’s the point. Compared to the Google+ version of photo storage that preceded it, Google Photos is not so much a social network as it is a private repository. If people are adding collaborative albums to their own personal photo libraries, the inclusion of likes and comments could make things messy.

To that end, it’d still be nice if Google included more controls over who has access to a given album. Although the owner of an album can toggle off sharing and collaboration at any time, users will still need to be careful about who gets access to shared links in the first place, as someone with the link could then share it with anyone.

Why this matters: Google is hardly the first company to tackle group photo albums, and collaboration is a long-standing feature of Google-owned Picasa. But the interesting thing about Google Photos is how it emphasizes ownership, letting collaborators easily save each others’ photos to their personal libraries without having to download anything. It’s an attempt to take some long-standing pain points out of cloud photo sharing, though the ability to freely copy and paste a link means it’s not quite as private as it could be.

Here’s a look at Walmart Pay in action

151210 walmart pay 1

This time last year, just as Apple Pay was being launched, tech enthusiasts were pouring scorn on rival payments system CurrentC, because it depends on barcodes rather than the newer NFC wireless technology picked up by Apple.

The jury is still out on CurrentC — a year on, it still hasn’t launched — but on Thursday Walmart announced a payment system of its own that uses barcodes, and it doesn’t look too clumsy.

In a video provided by the retailer, a customer starts an app on an Android phone with a credit card and a gift card stored in it. The customer selects Walmart Pay on a self-serve checkout machine and a barcode displays on the machine. He points his phone at the barcode and the two link-up, identifying the customer. To complete the identification, the phone needs some kind of network connection.

151210 walmart pay 3Walmart Pay being used in a company store

The customer then scans his purchases, hits the “I’m Done” button and the payment is processed.

The advantage of using barcodes is that the system can work on any smartphone, not just one with an NFC chip and antenna. Most high-end smartphones have the NFC hardware, but many older or cheaper phones do not. Shoppers then require only the Walmart app, which is available for iOS and Android.

Walmart Pay will launch in some stores this month and be in all stores by the middle of 2016. Only when it gets into the hands of consumers will a real test be possible.

While Walmart is one of the backers of CurrentC, it’s keen to point out that its system was developed in-house. It said it remains committed to CurrentC and will try out the technology when its ready.

Anonymous says it took down Trump Tower website

20151211 anonymous video still

The online activist group Anonymous said it took down the Trump Tower website on Friday after it warned presidential candidate Donald Trump about his statements on banning Muslims from entering the U.S.The site was unavailable during early afternoon, New York time, and according to media reports had been down for about an hour earlier in the day. Around 9 a.m. Friday there, the Anonymous Twitter account @YourAnonNews posted tweets saying the group had taken down the site.

The group also tweeted a link to a YouTube video in which it condemned Trump’s recent statements calling for a temporary ban on Muslims entering the U.S. Anonymous said Trump’s comments would help the Islamic State group recruit terrorists.

“Donald Trump, think twice before you speak anything,” the video said.

Trump Tower, a 68-story, mixed-use building in midtown Manhattan, is the flagship property of the Trump Organization, the international real-estate company that Donald Trump leads as chairman and president.

Anonymous frequently hacks websites in the name of political causes and says it is engaged in a cyberwar against the Islamic State. It claims to have taken down thousands of Twitter accounts that support the group.

Earlier this week, Trump said the U.S. should bar all Muslims from entering the U.S. as a measure to prevent terror attacks, “until our country’s representatives can figure out what’s going on.” The statements drew widespread condemnation in the U.S. and elsewhere but support from some right-leaning commentators. An NBC/Wall Street Journal poll found 57 percent of Americans opposed the idea.

Trump is the leading Republican candidate in the 2016 presidential election.

How to easily secure your web browsing with Hotspot Shield’s free, unlimited proxy

Data matrix networking connections system
Back in the spring we talked about TunnelBear for Chrome, an extension that adds an encrypted proxy to your browser. At the time, there weren’t many great options for Firefox, but since then Hotspot Shield rolled out its own free proxy add-ons for Firefox and Chrome.Encrypted proxies are a fast and lightweight alternative to virtual private networks (VPNs). They don’t offer the full breadth of protection that VPNs do since a proxy only protects your browser. But if all you want to do is view the U.S. Netflix catalog overseas or protect your browsing on a public Wi-Fi network, then an encrypted proxy should work just fine.

The best part about Hotspot Shield is that its proxy is free to use with unlimited bandwidth—TunnelBear on Chrome limits free users to 500 megabytes per month.

hotspotshield
Hotspot Shield VPN proxy for Firefox.

To get started on Firefox, download and install the add-on from Mozilla’s site. Once that’s done, a red Hotspot Shield logo should appear in your browser toolbar to the right of the address bar.

Just click on it, choose your virtual location and flip the on-off slider. Once that’s done, the Hotspot Shield logo should turn green. Now you’re ready to roll with HotSpot Shield on Firefox.

Although TunnelBear is my go-to VPN right now, I used to use Hotspot Shield quite a bit. Previously, Hotspot Shield’s free offering did not use browser add-ons and injected ads at the top of each browser tab when you landed on a new webpage. It was free, but all those ads really took a toll on performance so it’s great to see Hotspot Shield backing away from that.

If you want to use Hotspot Shield on Chrome, which is almost identical to the Firefox version, you can find it on the Chrome Web Store.