Regularization is used to find a solution that both fits the data and is sufficiently smooth, and thereby is very effective for designing and refining learning algorithms. But the influence of its exponent remains poorly understood. In particular, it is unclear how the exponent of the reproducing kernel Hilbert space~(RKHS) regularization term affects the accuracy and the efficiency of kernel-based learning algorithms. Here we consider regularized least squares regression (RLSR) with an RKHS regularization raised to the power of m, where m is a variable real exponent. We design an efficient algorithm for solving the associated minimization problem, we provide a theoretical analysis of its stability, and we compare its advantage with respect to computational complexity, speed of convergence and prediction accuracy to the classical kernel ridge regression algorithm where the regularization exponent m is fixed at 2. Our results show that the m-power RLSR problem can be solved efficiently, and support the suggestion that one can use a regularization term that grows significantly slower than the standard quadratic growth in the RKHS norm.

We describe a computationally efficient, stochastic graph-regularization technique that can be utilized for the semi-supervised training of deep neural networks in a parallel or distributed setting. We utilize a technique, first described in [13] for the construction of mini-batches for stochastic gradient descent (SGD) based on synthesized partitions of an affinity graph that are consistent with the graph structure, but also preserve enough stochasticity for convergence of SGD to good local minima. We show how our technique allows a graph-based semi-supervised loss function to be decomposed into a sum over objectives, facilitating data parallelism for scalable training of machine learning models. Empirical results indicate that our method significantly improves classification accuracy compared to the fully-supervised case when the fraction of labeled data is low, and in the parallel case, achieves significant speed-up in terms of wall-clock time to convergence. We show the results for both sequential and distributed-memory semi-supervised DNN training on a speech corpus.

T ensors have a rich history, stretching over almost a century, and touching upon numerous disciplines; but they have only recently become ubiquitous in signal and data analytics at the confluence of signal processing, statistics, data mining and machine learning. This overview article aims to provide a good starting point for researchers and practitioners interested in learning about and working with tensors. As such, it focuses on fundamentals and motivation (using various application examples), aiming to strike an appropriate balance of breadth and depth that will enable someone having taken first graduate courses in matrix algebra and probability to get started doing research and/or developing tensor algorithms and software. Some background in applied optimization is useful but not strictly required. The material covered includes tensor rank and rank decomposition; basic tensor factorization models and their relationships and properties (including fairly good coverage of identifiability); broad coverage of algorithms ranging from alternating optimization to stochastic gradient; statistical performance analysis; and applications ranging from source separation to collaborative filtering, mixture and topic modeling, classification, and multilinear subspace learning. Index Terms --T ensor decomposition, tensor factorization, rank, canonical polyadic decomposition (CPD), parallel factor analysis (PARAF AC), T ucker model, higher-order singular value decomposition (HOSVD), multilinear singular value decomposition (MLSVD), uniqueness, NPhard problems, alternating optimization, alternating direction method of multipliers, gradient descent, Gauss-Newton, stochastic gradient, Cram er-Rao bound, communications, source separation, harmonic retrieval, speech separation, collaborative filtering, mixture modeling, topic modeling, classification, subspace learning. N.D. Sidiropoulos, X. Fu, and K. Huang are with the ECE Department, University of Minnesota, Minneapolis, USA; email: (nikos,xfu,huang663)@umn.edu .

General introduction Network and link analysis is a highly studied field, subject of much recent work in various areas of science: applied mathematics, computer science, social science, physics, chemistry, pattern recognition, applied statistics, data mining & machine learning, to name a few [4, 20, 30, 56, 61, 73, 96, 101]. Within this context, one key issue is the proper quantification of the structural relatedness between nodes of a network by taking both direct and indirect connections into account. This problem is faced in all disciplines involving networks in various types of problems such as link prediction, community detection, node classification, and network visualization to name a few popular ones. Preprint submitted to Elsevier January 2, 2018 The main contribution of this paper is in presenting in detail the bag-ofpaths (BoP) framework and defining relatedness as well as distance measures between nodes from this framework. The BoP builds on and extends previous work dedicated to the exploratory analysis of network data [54, 53, 67, 104]. The introduced distances are constructed to capture the global structure of the graph by using paths on the graph as a building block. In addition to relatedness/distance measures, various other quantities of interest can be derived within the probabilistic BoP framework in a principled way, such as betweenness measures quantifying to which extent a node is in between two sets of nodes [60], extensions of the modularity criterion for, e.g., community detection [26], measures capturing the criticality of the nodes or robustness of the network, graph cuts based on BoP probabilities, and so on.

Not long ago, news of a massive asteroid hurtling toward Earth would be a death sentence that humans, like the dinosaurs before us, would simply have to accept. But as near-Earth object detection has gotten more sophisticated, a new field called Planetary Defense has risen to meet the challenge of defending Earth from asteroids. However, speaking at the annual meeting of the American Geophysical Union, Joseph Nuth, a researcher with NASA's Goddard Space Flight Center, warned that while asteroids and comets capable of resulting in mass extinctions are rare – as in, every-60-million-years rare – even small objects from space can cause a great deal of damage and the planet's scientists are are still woefully unprepared to deal with the threat. "It is really imperative that we reduce our reaction time," Dr. Nuth said at the meeting. "The biggest problem, basically, is there's not ... a lot we can do about it at the moment."

After eight years and two million miles, Google officially announced the launch of a new company dedicated solely to its self-driving cars. "We believe that this technology can begin to reshape some of the ten trillion miles that motor vehicles travel around the world every year with safer, more efficient and more accessible forms of transport," the company said in a blog post on its website Tuesday. Waymo, which stands for "a new way forward in mobility," aims to make roads safer by ending drunk and distracted driving and make cars accessible for people with limited mobility. Google completed the world's first fully autonomous trip in 2015 when a legally blind man drove through suburban Texas, relying solely on the car to pilot. "We are a self-driving technology company. We've been really clear that we're not a car company, although there's sometimes some confusion on that point," said CEO John Krafcik.

Finding customer service help online can be a pain. Filtering through a knowledge base to find the right answer to your question can be an exercise in fighting with nested frequently asked questions documents. Microsoft is aiming to help by making it easier for companies to create intelligent bots that can answer common questions. The QnA Maker, launched in beta on Tuesday, will let users train an automated conversation partner on existing frequently-asked-questions content. After that information is fed in, the service will create a bot that will respond to customer questions with the content from the knowledge base.

Speaking at Mobey Day in Toronto, IBM associate partner Pavel Abdur-Rahman moved to cool some of the hype surrounding AI, arguing that it is not going to be seen in a true sense until quantum computing takes off more than a decade down the line. However, the data scientist was keen to show off the capabilities of Big Blue's cognitive computing platform, Watson, and what this means for the financial services industry. Abdur-Rahman ran through seven ways in which Watson can help banks, ranging from the simple, such as a chatbot acting as a virtual assistant answering basic customer queries, to the "quite complex", including looking for patterns in unstructured data to find early signals of opportunities and risks for asset managers building portfolios. Watson caused a stir in 2011 when it beat two human competitors to claim the $1 million prize on the US quiz show Jeopardy. Since then several banks, including Standard Bank, USAA, DBS Bank, ANZ and Royal Bank of Canada have been attempting to take the platform out of the TV studio and into the real world, harnessing its power for a variety of tasks.

From advanced driver assistance systems and security solutions to concepts for cleaner drives and holistic connectivity, the list of pioneering products from Continental is expanding as the automotive industry continues its rapid shift towards software and mobility services. To demonstrate its interpretation of the digitalization of the automotive world, the international technology company will be showcasing a collection of products and services at CES 2017 on January 5 - 8 in Las Vegas. "The entire automotive industry is undergoing a revolution and we at Continental are playing an active role in shaping it. On the basis of our solid expertise in systems integration and our years of experience in the field of vehicle connectivity, we are constantly adapting our products in line with the latest market trends," said Helmut Matschi, member of the Executive Board at Continental and head of the Interior Division. "For example, we are developing the building blocks of new mobility solutions for our customers and helping to shape the mobility of tomorrow together. At the CES, we will be showcasing the very latest solutions that will take us into the future."

"Most current advanced driver assistance systems based on radar and cameras are not capable of accurately detecting and classifying objects – such as cars, pedestrians or bicycles – at a level required for autonomous driving," said Sachin Lawande, president and CEO of Visteon, a leading global cockpit electronics supplier. "We need to achieve virtually 100 percent accuracy for autonomous driving, which will require innovative solutions based on deep machine learning technology. Our Silicon Valley team, with its focus on machine learning software development, will be a critical part of our autonomous driving technology initiative." Visteon's recently opened facility in the heart of Silicon Valley will house a team of engineers specializing in artificial intelligence and machine learning. The center is located close to the West Coast offices of various automakers and tech companies, as well as Stanford University and the University of California, Berkeley – two of the leading universities for artificial intelligence and deep learning in the U.S. In addition to leading Visteon's artificial intelligence efforts, the Silicon Valley office will play a key role in delivering control systems, localization and vision processing – interpreting live camera data and converting it to information required for autonomous driving.