Microsoft has reorganized several disparate projects and programs into an entirely new Artificial Intelligence group. It will be the fourth major division after Windows, Office and cloud, reallocating over 5,000 computer scientists and engineers under its umbrella. The shift shows how much unified effort the tech giant believes the field needs, as well as internally standardized AI tech they can more easily integrate into customer products. Microsoft Research chief Harry Shum will head the new AI division, according to GeekWire. In addition to his old department, the new group will include products like Cortana and Bing with the Ambient Computing and Robotics teams, as well as the company's Information Platform Group. Collecting them all under one banner suggests how standardized they want their AI tech to be between disparate programs.

If you spend your work days toiling in Google's productivity apps, the first thing you might notice today is that Google for Work is now called "G Suite". Once you get past the new label, you might also notice a slew of smart updates across the board that ought to save you time and keep your workflow moving. First up: Docs, Sheets and Slides got a new "Explore" feature that uses natural language search to help you research reports, organize data or design better looking presentations. In each of the main apps, an Explore button brings up a new sidebar with contextual options based on the app you're using. In Docs, this means Explore will search and suggest images, web links or other Drive documents that appear relevant to the content you're writing.

There may now be an easier way to implement advanced machine learning models in your projects. Google has opened up its Cloud Machine Learning to all businesses in a public beta, after a few months of testing it in private alpha. The tool makes it easier to train models at a much faster rate, and is integrated with the Google Cloud Platform. This has applications for businesses in areas such as customer support (learning how to automate responses to a variety of queries and complaints) or any kind of repetition-heavy task. In a blog post, Google described how its customer Airbus Defense and Space used the tool to automate the detection and correction of satellite images that contain imperfections such as cloud formations.

Americans want to stay in control of their cars, a new study finds. According to a study by Kelley Blue Book, 80 percent of Americans say people should always have the option to drive themselves. This study comes just a week after the Department of Transportation released regulatory guidelines for self-driving vehicles. And it comes comes as car companies are spending billions to advance the technology. But despite the push toward autonomous cars, consumers remain unconvinced.

You're so hard to explore. Sometimes you bombard spacecrafts with hurtling rocks and deadly cosmic rays, and other times you're so empty you don't give astronauts a darn thing to hold on to. But while scientists haven't quite figured out how to keep radiation at bay, the scientists at NASA's Jet Propulsion Laboratory--specifically, its Planetary Robotics Laboratory--are building machines that can get a grip on the most difficult surfaces astronauts will find out there. Adhesion-wise, space presents a couple problems. First, robots typically struggle with uneven surfaces, let alone the kind of cliffs and crags you see on Mars.

IBM said Thursday it plans to acquire compliance consulting firm Promontory Financial Group to bring more financial regulatory expertise to Watson's cognitive computing platform. Promontory is a global consulting operation with an aim of helping banks manage the ever-increasing regulation and risk management requirements in the financial sector. With that in mind, IBM wants to use the industry expertise of Promontory's workforce -- which is made up of ex-regulators and banking executives -- to teach Watson all about regulation, risk and compliance. IBM is also using the deal to create a new subsidiary called Watson Financial Services, which will build cognitive tools for things things like tracking regulatory obligations, financial risk modeling, surveillance, anti-money laundering detection systems. "This is a workload ideally suited for Watson's cognitive capabilities intended to allow financial institutions to absorb the regulatory changes, understand their obligations, and close gaps in systems and practices to address compliance requirements more quickly and efficiently," IBM said in a press release.

Diane Greene, Google's senior vice president of cloud, announced Thursday that Google's enterprise business is officially rebranding as Google Cloud. "We're the full power of Google in the Cloud," said Greene at a San Francisco, Calif., event. Initially, "we were sort of Google for Work, Google Cloud Platform, Google Apps, Enterprise," she said. The enterprise business, she continued, offers a "uniquely Google broad set of technologies, solutions, products," from Android and secure Chromebooks to the Google Cloud Platform and its machine learning capabilities. Since Greene was tapped to join Google late last year, the company has made a concerted effort to bring on the corporate clients that Google needs to compete in the cloud space against Amazon Web Services. Since joining, Greene said she's talked to more than 200 of Google's partners and customers.

Google on Thursday announced a series of enterprise tools powered by machine learning. "Our goal is to democratize machine learning," said Urs Hölzle, senior vice president of technical infrastructure for cloud, at a San Francisco, Calif., event Thursday. He announced the introduction of Cloud Machine Learning in beta. Security is everyone's responsibility, not just those with the cash to upgrade. It is a NoOps machine learning solution, making it easy for businesses to run and train machine learning models.

Google's end-user applications have always been useful, and they've been reasonably well-integrated. Now, with Diane Greene, founder and former CEO of VMware, calling Google's cloud shots, things have gotten more closely aligned. In G Suite, Google Apps for Work hasn't just been renamed. With the new G Suite Hangouts you'll be able to automatically work out what time works best for everyone to meet. True, the same apps will be in G Suite as those in Google Apps for Work.

Tl;dr - has this idea about neural network architecture, similar to autoencoders, been developed already? Correct me if I'm wrong about anything, but this is my understanding so far: Autoencoders purposefully have an information bottleneck in the middle, and this bottleneck is what forces the network to learn high level representations of the input data. Otherwise, without the bottleneck, the network may discover that the optimal connection is one that is roughly equivalent to directly mapping each input to its corresponding output. However, because that bottleneck exists, it is very unlikely that an autoencoder could perform a perfect reconstruction of the input. Contrast this with other encoding, like DCT as used by JPEG images.