While variational dropout approaches have been shown to be effective for network sparsification, they are still suboptimal in the sense that they set the dropout rate for each neuron without consideration of the input data. With such input-independent dropout, each neuron is evolved to be generic across inputs, which makes it difficult to sparsify networks without accuracy loss. To overcome this limitation, we propose adaptive variational dropout whose probabilities are drawn from sparsity-inducing beta-Bernoulli prior. It allows each neuron to be evolved either to be generic or specific for certain inputs, or dropped altogether. Such input-adaptive sparsity- inducing dropout allows the resulting network to tolerate larger degree of sparsity without losing its expressive power by removing redundancies among features. We validate our dependent variational beta-Bernoulli dropout on multiple public datasets, on which it obtains significantly more compact networks than baseline methods, with consistent accuracy improvements over the base networks.

Deep learning is the mainstream technique for many machine learning tasks, including image recognition, machine translation, speech recognition, and so on. It has outperformed conventional methods in various fields and achieved great successes. Unfortunately, the understanding on how it works remains unclear. It has the central importance to lay down the theoretic foundation for deep learning. In this work, we give a geometric view to understand deep learning: we show that the fundamental principle attributing to the success is the manifold structure in data, namely natural high dimensional data concentrates close to a low-dimensional manifold, deep learning learns the manifold and the probability distribution on it. We further introduce the concepts of rectified linear complexity for deep neural network measuring its learning capability, rectified linear complexity of an embedding manifold describing the difficulty to be learned. Then we show for any deep neural network with fixed architecture, there exists a manifold that cannot be learned by the network. Finally, we propose to apply optimal mass transportation theory to control the probability distribution in the latent space.

A parasitic fungus that turns carpenter ants into'zombies' is learning to adapt to the changing climate. The fungus is known to infect carpenter ants and cause them to clamp onto hanging vegetation, where they dangle for months to spew spores. But, researchers have found that zombie ants don't always bite onto the same part of the plant. According to a new study, this is because the choice of leaves or twigs boils down to the local climate – and, as conditions continue to change, the fungi have been forced to adapt. A parasitic fungus that turns carpenter ants into'zombies' is learning to adapt to the changing climate.

Businesses need to better fund AI cybersecurity to combat the growing threat of artificially intelligent (AI) cyberattacks, according to the CTO of the Cyber Security Research Center at Ben-Gurion University, Dudu Mimran. Speaking at the OECD Forum, Mimran predicted that AI would become a more common weapon in the arsenal of hackers. The senior telecommunications and cybersecurity expert argued that this is partly because using AI would be "cheaper, faster and smarter" for hackers. AI-driven cyberattacks could take many forms, including phishing, identity theft and denial-of-service attacks. According to Natan Bandler, CEO and co-founder of Cy-oT, the two main areas in which AI is being used to carry out attacks are as a "tool to find exploits" and for auto hacking to "map existing exploits and weaknesses."

Above the Trend Line: your industry rumor central is a recurring feature of insideBIGDATA. In this column, we present a variety of short time-critical news items grouped by category such as people movements, funding news, financial results, industry alignments, rumors and general scuttlebutt floating around the big data, data science and machine learning industries including behind-the-scenes anecdotes and curious buzz. Our intent is to provide you a one-stop source of late-breaking news to help you keep abreast of this fast-paced ecosystem. We're working hard on your behalf with our extensive vendor network to give you all the latest happenings. Be sure to Tweet Above the Trend Line articles using the hashtag: #abovethetrendline.

Amazon has led in smart speaker sales and adoption since it pioneered the category with the Echo in 2014. But for the first time, someone else has topped the smart speaker sales charts. According to data from technology and smartphone analysts with Canalys, Google beat out Amazon in first quarter smart speaker shipments this year. Google shipped 3.2 million Google Home and Home mini products in the first quarter of this year, while Amazon shipped only 2.8 million. Google has been inching closer to Amazon's sales numbers, largely aided by a gangbuster holiday season in which it grabbed 40 percent of U.S. smart speaker sales.

What does digital transformation really mean in retail? It is simply transferring an old, obsolete product-centered model that has been followed for so many years to a customer-centered model. This is somewhat easy to say, but again, what does it specifically mean for the retail industry? Instead of focusing on a supply chain view – cheaper purchase, higher sales, and optimizing everything in all of these – retailers should focus more on the digital value chain view. They must be able to collect data regarding their customers, products and locations, turn those information into useful insights, and then turn those insights into planning and action.

Russia isn't going to sit on the sidelines while American companies like Amazon and Google conquer the smart speaker world. Search giant Yandex has launched the Yandex.Station, billed both as Russia's first smart speaker and the company's first hardware product. The device uses Yandex's in-house voice assistant, Alice, to handle upwards of 4,000 skills like checking traffic, ordering pizza or finding flights. However, its real specialty is video -- it's an unusual bridge between audio-only speakers and smart displays. The Yandex.Station has an HDMI output that lets you stream videos directly to your TV, whether it's from Yandex's own KinoPoisk or alternatives like Ivi and Amediateka.

Intuitive Surgical (ISRG) started with a very simple plan. That is, to make surgery less invasive with surgical robots. For laypeople in 1995, the concept was science fiction. At the time, researchers at the Stanford Research Institute had been kicking around the idea for years. That's because the U.S. Army had hired them in the late '80s to make remote battlefield surgery feasible.

Uber has launched a'panic button' allowing passengers to call 911 directly from its app during a ride. The emergency button is part of a new'safety center' menu designed to help passengers if something goes wrong during their trip. It is hoped the feature will help allay concerns over the safety of Uber. The emergency button is located in a new'safety center' menu that is easily accessible from the app's home screen, giving riders a quick way to contact first responders in the event that something goes wrong during their trip. To dial 911, riders will need to swipe up on the safety center icon, and then tap '911 assistance.'