The computation and storage requirements for Deep Neural Networks (DNNs) are usually high. This issue limits their deployability on ubiquitous computing devices such as smart phones, wearables and autonomous drones. In this paper, we propose ternary neural networks (TNNs) in order to make deep learning more resource-efficient. We train these TNNs using a teacher-student approach based on a novel, layer-wise greedy methodology. Thanks to our two-stage training procedure, the teacher network is still able to use state-of-the-art methods such as dropout and batch normalization to increase accuracy and reduce training time. Using only ternary weights and activations, the student ternary network learns to mimic the behavior of its teacher network without using any multiplication. Unlike its -1,1 binary counterparts, a ternary neural network inherently prunes the smaller weights by setting them to zero during training. This makes them sparser and thus more energy-efficient. We design a purpose-built hardware architecture for TNNs and implement it on FPGA and ASIC. We evaluate TNNs on several benchmark datasets and demonstrate up to 3.1x better energy efficiency with respect to the state of the art while also improving accuracy.

Existing approaches to resource allocation for nowadays stochastic networks are challenged to meet fast convergence and tolerable delay requirements. The present paper leverages online learning advances to facilitate stochastic resource allocation tasks. By recognizing the central role of Lagrange multipliers, the underlying constrained optimization problem is formulated as a machine learning task involving both training and operational modes, with the goal of learning the sought multipliers in a fast and efficient manner. To this end, an order-optimal offline learning approach is developed first for batch training, and it is then generalized to the online setting with a procedure termed learn-and-adapt. The novel resource allocation protocol permeates benefits of stochastic approximation and statistical learning to obtain low-complexity online updates with learning errors close to the statistical accuracy limits, while still preserving adaptation performance, which in the stochastic network optimization context guarantees queue stability. Analysis and simulated tests demonstrate that the proposed data-driven approach improves the delay and convergence performance of existing resource allocation schemes.

In this paper, we propose TopicRNN, a recurrent neural network (RNN)-based language model designed to directly capture the global semantic meaning relating words in a document via latent topics. Because of their sequential nature, RNNs are good at capturing the local structure of a word sequence - both semantic and syntactic - but might face difficulty remembering long-range dependencies. Intuitively, these long-range dependencies are of semantic nature. In contrast, latent topic models are able to capture the global underlying semantic structure of a document but do not account for word ordering. The proposed TopicRNN model integrates the merits of RNNs and latent topic models: it captures local (syntactic) dependencies using an RNN and global (semantic) dependencies using latent topics. Unlike previous work on contextual RNN language modeling, our model is learned end-to-end. Empirical results on word prediction show that TopicRNN outperforms existing contextual RNN baselines. In addition, TopicRNN can be used as an unsupervised feature extractor for documents. We do this for sentiment analysis on the IMDB movie review dataset and report an error rate of $6.28\%$. This is comparable to the state-of-the-art $5.91\%$ resulting from a semi-supervised approach. Finally, TopicRNN also yields sensible topics, making it a useful alternative to document models such as latent Dirichlet allocation.

At about lunchtime on an unseasonably warm February day, a small commercial drone hovered alongside Highlight Tower; a striking, angular glass block soaring 126m over a suburban Autobahn on the outskirts of Munich, with equally striking views. This email address is already registered. By submitting my Email address I confirm that I have read and accepted the Terms of Use and Declaration of Consent. By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers. You also agree that your personal information may be transferred and processed in the United States, and that you have read and agree to the Terms of Use and the Privacy Policy.

MIT's Kate Darling, who writes the rules of human-robot interaction, says an AI-enabled apocalypse should be the least of our concerns. A s a law student in Switzerland, Kate Darling's interest in robots was just a hobby. She had purchased a PLEO robot dinosaur that was designed to respond to human contact emotionally and act independently. "It really struck me that I responded to the cues the robot was giving me, even though I knew exactly how the toy worked," Darling says. "I knew where all the motors were and how it worked, and why it would cry when you held it up by the tail, but I was just so compelled to comfort it and make it stop crying."

Researchers at Microsoft and Cambridge University have built a highly sophisticated computer called DeepCoder that can now allow machines to write their own programs. This is aimed to make job easier for users who don't know programming languages well enough to use them efficiently, or even help people having no experience in writing simple coding programs. "All of a sudden people could be so much more productive," says Armando Solar-Lezama at the Massachusetts Institute of Technology, who was not involved in the work. "They could build systems that it [would be]impossible to build before." The paper, DeepCoder: Learning to Write Programs, is a basic system with some limitations according to the researchers.

Stock predictions made by machine learning are being deployed by a select group of hedge funds that are betting that the technology used to make facial recognition systems can also beat human investors in the market. Computers have been used in the stock market for decades to outrun human traders because of their ability to make thousands of trades a second. More recently, algorithmic trading has programmed computers to buy or sell stocks the instant certain criteria is met, such as when a stock suddenly becomes cheaper in one market than in another -- a trade known as arbitrage. Machine learning, an offshoot of studies into artificial intelligence, takes the stock trading process a giant step forward. Pouring over millions of data points from newspapers to TV shows, these AI programs actually learn and improve their stock predictions without human interaction.

Editor's Note: This article first appeared in Spektrum der Wissenschaft, Scientific American's sister publication, as "Digitale Demokratie statt Datendiktatur." "Enlightenment is man's emergence from his self-imposed immaturity. Immaturity is the inability to use one's understanding without guidance from another." The digital revolution is in full swing. How will it change our world? The amount of data we produce doubles every year. In other words: in 2016 we produced as much data as in the entire history of humankind through 2015. Every minute we produce hundreds of thousands of Google searches and Facebook posts. These contain information that reveals how we think and feel. Soon, the things around us, possibly even our clothing, also will be connected with the Internet. It is estimated that in 10 years' time there will be 150 billion networked measuring sensors, 20 times more than people on Earth. Then, the amount of data will double every 12 hours. Many companies are already trying to turn this Big Data into Big Money. Everything will become intelligent; soon we will not only have smart phones, but also smart homes, smart factories and smart cities. Should we also expect these developments to result in smart nations and a smarter planet? The field of artificial intelligence is, indeed, making breathtaking advances. In particular, it is contributing to the automation of data analysis. Artificial intelligence is no longer programmed line by line, but is now capable of learning, thereby continuously developing itself. Recently, Google's DeepMind algorithm taught itself how to win 49 Atari games. Algorithms can now recognize handwritten language and patterns almost as well as humans and even complete some tasks better than them. They are able to describe the contents of photos and videos. Today 70% of all financial transactions are performed by algorithms. News content is, in part, automatically generated. This all has radical economic consequences: in the coming 10 to 20 years around half of today's jobs will be threatened by algorithms. It can be expected that supercomputers will soon surpass human capabilities in almost all areas--somewhere between 2020 and 2060.

Octopus-like robots are plucking strawberries in Spain, in the US machines are vacuuming apples off the trees, and in the UK they are feeding and milking cows. Robots are taking over fields around the world, and last week food and rural affairs secretary Andrea Leadsom suggested they could help replace the thousands of EU workers who currently help put food on British tables. And it is not just Brexit that is forcing the agricultural industry to embrace the next phase of mechanisation. Farmers are already having to rethink their operations in the face of higher minimum pay – mainly a result of the national living wage for over-25s, which came into effect last year. Robotic milking machines, in which cows queue up to milk themselves, are now mainstream, while systems tat automatically feed or track the health of livestock are on the rise.

After a delayed launch and one aborted delivery attempt, SpaceX's caution paid off Thursday when its Dragon capsule stuffed full of food, equipment, and experiments successfully docked with the International Space Station (ISS). Now on its 10th re-supply mission, the private space company has become an essential part of the supply lines supporting an increasingly intricate space operation. After a GPS error scuttled its first docking attempt Wednesday, the Dragon capsule smoothly slipped close enough to the ISS for the space station's robotic arm to snag the craft early Thursday morning, along with the 5,500 pounds of goodies on board. "Looks like we've got a great capture," radioed space station commander Shane Kimbrough. In addition to a much needed food refresh, the capsule also contains more than 250 science experiments.