Detecting and classifying targets in video streams from surveillance cameras is a cumbersome, error-prone and expensive task. Often, the incurred costs are prohibitive for real-time monitoring. This leads to data being stored locally or transmitted to a central storage site for post-incident examination. The required communication links and archiving of the video data are still expensive and this setup excludes preemptive actions to respond to imminent threats. An effective way to overcome these limitations is to build a smart camera that transmits alerts when relevant video sequences are detected. Deep neural networks (DNNs) have come to outperform humans in visual classifications tasks. The concept of DNNs and Convolutional Networks (ConvNets) can easily be extended to make use of higher-dimensional input data such as multispectral data. We explore this opportunity in terms of achievable accuracy and required computational effort. To analyze the precision of DNNs for scene labeling in an urban surveillance scenario we have created a dataset with 8 classes obtained in a field experiment. We combine an RGB camera with a 25-channel VIS-NIR snapshot sensor to assess the potential of multispectral image data for target classification. We evaluate several new DNNs, showing that the spectral information fused together with the RGB frames can be used to improve the accuracy of the system or to achieve similar accuracy with a 3x smaller computation effort. We achieve a very high per-pixel accuracy of 99.1%. Even for scarcely occurring, but particularly interesting classes, such as cars, 75% of the pixels are labeled correctly with errors occurring only around the border of the objects. This high accuracy was obtained with a training set of only 30 labeled images, paving the way for fast adaptation to various application scenarios.

Facebook started to slowly introduce its new deep learning platform called Caffe2go, which lets users capture and transfer video effects in real time using iOS or Android smartphones. Though the effects are pretty cool, the tech behind Caffe2Go is really interesting. Deep learning requires content to "be sent off to data centers for processing on big-compute servers," Facebook wrote, but with Caffe2go, the processing can be done from "the palm of your hand." The new platform is part of Facebook's AI effort that includes the Lumos app used to take out images that violate its standards. Google released Tensorflow framework to the open source community and Microsoft made its Cognitive toolkit available to developers.

Decades-old discoveries are now electrifying the computing industry and will soon transform corporate America. Over the past four years, readers have doubtlessly noticed quantum leaps in the quality of a wide range of everyday technologies. Most obviously, the speech-recognition functions on our smartphones work much better than they used to. When we use a voice command to call our spouses, we reach them now. We aren't connected to Amtrak or an angry ex.

Do you know why you can't hear the ugly ahem sounds on the podcast Data Science at Home? The ahem detector is a deep convolutional neural network trained on transformed audio signals to recognize ahem sounds. The network has been trained to detect such signals on the episodes of Data Science at Home, the podcast about data science at worldofpiggy.com/podcast But before proceeding, some concepts should be clarified. While the detector works for the aforementioned audio files, it can be generalized to any other audio input, provided enough data are available.

Not only are online retailers competing with other online stores and brick-and-mortar locations, but also the overall noise that is the Internet. We live in a world where consumer attention span is getting shorter and shorter: 40 percent of people abandon a website that takes more than three seconds to load, and the average shopping cart is abandoned more than 68 percent of the time. I'm hard pressed to find an ecommerce site that is not constantly scrambling to engage more and drive more sales. Technology is finally helping with those efforts in a big way. Artificial intelligence (AI), which has demonstrated its value in industries like marketing, healthcare and finance, is now making a splash in online commerce.

Google, Amazon, and Facebook can magically recognize images and voices, thanks to superfast servers equipped with GPUs in their mega data centers. But not all companies can afford that level of resources for deep learning, so they turn to cloud services, where servers in remote data centers do the heavy lifting. Microsoft has made such cloud services trendy with Azure and is one of the few companies offering remote servers with GPUs, which excel in machine-learning tasks. But Azure uses older Nvidia GPUs, and it now has competition from Nimbix, which offers a cloud service with faster GPUs based on the Nvidia's latest Pascal architecture. After renting time on the cloud service, customers get a virtual machine with access to bare-metal server hardware.

Google cut power usage in its data centers by several percentage points earlier this year by trusting artificially intelligent software derived from 1980s-era Atari video games. And in the years to come, the Internet giant not only could save much more electricity, but also solve far larger problems by taking on a much more complex video game. Research scientists at Google's DeepMind unit announced Friday they are developing a computer program that reads data about Blizzard Entertainment's "StarCraft II" games and learns how to play on its own. The software would have to figure out how to split its attention between micromanagement and long-term strategic decisions. It's that maneuvering that could deliver big breakthroughs.

Crunching vast tracts of data is a growing task for defense, intelligence and security agencies. They need analysis, fast, of what's going on in the air and on the ground to assess battlefields, secure environments, and decide when and how to deploy people or humanitarian aid. Artificial intelligence may be the key to digesting the barrage of data from multiple sources. To unlock insights from this data, agencies are increasingly turning to GPU-powered deep learning, with algorithms that can identify relevant content and patterns in raw data at machine speed. The GPU is the engine of modern AI, NVIDIA solution architect Jon Barker Barker told a broad audience from the defense, intelligence and homeland security communities at the recent annual GEOINT Symposium.

It's now a hired gun for thousands of companies in at least 20 industries. David Kenny took the helm of IBM's Watson Group ibm in February, after Big Blue acquired The Weather Company, where Kenny had served as CEO. In the months since then, the Watson business has grown dramatically, with well over 100,000 developers worldwide now working with more than three dozen Watson application program interfaces (APIs). Fortune Deputy Editor Clifton Leaf caught up with Kenny in mid-October, when IBM Watson's General Manager was in San Francisco, getting ready to open Watson West--the AI system's newest business outpost--and to launch the company's second World of Watson conference, a gathering of its burgeoning ecosystem of partners and users, in Las Vegas on Oct. 24. KENNY: Deep learning is a subset of machine learning, which essentially is a set of algorithms. Deep-learning uses more advanced things like convolutional neural networks, which basically means you can look at things more deeply into more layers. Machine learning could work, for example, when it came to reading text.

Many have written about how deep learning is taking over the world and why that is important; I cannot echo them enough. Playing with deep learning is the closest I've ever felt to being a magician, and it's become clear to me that every (great) piece of software will be powered by deep learning within the next 3 years. However, deep learning isn't mainstream yet, so I thought I'd share work by some very talented contributors, in the hopes to bring it just that little bit closer. Quick note: I've started a weekly email newsletter covering all things deep learning and self-driving cars. I call it Transmission, sign up today!