In fact, it's been a while since the song-identifying software squeezed its way into the iOS App Store's top 10. So, why has Apple confirmed it is "combining" its business with that of the smaller London company? It has not revealed the price it is paying, but the sum is rumoured to be as much as $400m (£300m), which would make it one of Apple's most expensive takeovers to date. The US technology giant also hasn't disclosed its motivations beyond saying that it has "exciting plans in store". But there are several reasons the deal may have appealed.

If you thought Sphero's original BB-8 app-enabled robot was a blast back in 2015, just wait'til you see their take on Star Wars' most beloved droid: R2-D2. MSRP; same price on Amazon and the Apple store) is here in advance of Star Wars: The Last Jedi, which continues the sci-fi saga's main storyline, and it's just as immediately impressive as BB-8. It'll cruise around your home with ease using the iPhone/iPad companion app, the head rotates and reacts just like the real droid in the movie, and it even has movie-perfect sound effects and little animation routines. R2-D2 is not alone this time around, either, as the First Order BB-9E ($149.95 MSRP; same price on Amazon and the Apple store) droid from the new movie--essentially an evil version of BB-8, it seems--has joined the Sphero crew.

The other day I received a voice notification that my Amazon order for Vitamin D pills (getting ready for a Portland winter) had shipped and was en route to my home. It was a simple gesture that was also communicated via email. But for someone who sends billions of notifications a day, the voice alert got my juices flowing because it represented an exciting future that's starting to come into view. After years of science fiction exposing us to the idea of voice-powered apparatus, we finally live in an era where smart speakers such as Amazon Echo, Google Home and other devices (like Apple's forthcoming HomePod) are at our beck and call to deliver news, music and movies in one moment and then, a few seconds later, change the temperature of our living room. And get ready for more notifications.

It hasn't been an easy couple of years for algorithms. Increasingly populated with content decried as'fake news', 'clickbait', today's highly personalized social media feeds are coming under increasing fire for being filter bubbles, culminating in Mark Zuckerberg's highly public apology to Facebook's users at the start of this month. These shifts in media mirror concurrent developments in spheres as diverse as customer service and support, financial trading, healthcare, and more. Today, though, tech is fighting back – thanks to AI. After partnering with Automated Insights in 2014 – a natural language generation start-up – the Associated Press became one of the earliest adopters of AI within the media space.

The era of Artificial Intelligence (AI) is upon us. More and more, computer systems will become able to perform tasks that previously required human intelligence, such as decision making, visual perception, and speech and language recognition. Marketing, like most other fields, will feel AI's impact in several areas, including database marketing techniques, search queries and search engine optimization (SEO), personalization, predictive customer service, sales forecasting, customer segmentation, pricing, and many others. Act-On recently asked several marketing experts and analysts for their assessments of the pros and cons of using Artificial Intelligence in marketing and what they foresee for the future. Question: Will Artificial Intelligence (AI) be marketing's friend or foe? How do you see the two co-existing as AI becomes more mainstream?

If you make a purchase by clicking one of our links, we may earn a small share of the revenue. However, our picks and opinions are independent from USA TODAY's newsroom and any business incentives. Amazon's 12 Days of Deals continues on with a slew of offers for "music lovers, avid readers, and film buffs" and we've cherry-picked the best offers on the best products from the lineup. For the person you want to splurge on, tech is just the ticket. And whether they're the type who always has music playing, prefers reading to TV, or is obsessed with the latest technology, one of these things is sure to be the perfect fit.

People used to dream about robot secretaries. Futurists imagined a world where computers obeyed our every command. When the Amazon Echo hit the market, that dream became a reality: Alexa was obedient, personable and all-knowing. She could carry out a myriad of basic tasks, with a personality as professional and unflappable as a human assistant. This year, Echo is more popular than ever, and despite competition from Google, Amazon still dominates 75 percent of the virtual assistant market.

Previous works have used deep learning. They used regression to predict the colour of each pixel. This, however, produces fairly bland and dull results. Previous works used Mean Squared Error (MSE) as the loss function to train the model. The authors noted that MSE will try to'average' out the colors in order to get the least average error, which will result in a bland look.

More than 20 announcements flowed at AWS re:Invent today, including updates on machine learning, databases and the Internet of Things (IoT) – not to mention a couple of big-name customer wins. The theme for Amazon Web Services CEO Andy Jassy was around builders. Unlike the previous year's keynote, where it was about superpowers, this presentation had a slightly more mundane – and musical – approach. Technology builders are like musicians, Jassy explained. Some guitarists may play acoustic a lot of the time, but require some parts be played electric.

Hi, despite all the alchemy which Batch Norm does behind the covariate shift, I understand it simply as normalization layer which tries to keep all activation within some prior distribution. This is especially helpful at the beginning of the training process where badly chosen initialization may lead to vanishing or exploding signal in the network. Once the network is trained the batch norm in test time uses moving averages of estimated mean and variance of training population i.e. it applies simple linear transformation. Did anyone try to compare estimated values of mean and variance, with those computes from whole training set (i.e. I'm wonder how this would affect test accuracy.