First steps towards a mathematical theory of deep convolutional neural networks for feature extraction were made---for the continuous-time case---in Mallat, 2012, and Wiatowski and B\"olcskei, 2015. This paper considers the discrete case, introduces new convolutional neural network architectures, and proposes a mathematical framework for their analysis. Specifically, we establish deformation and translation sensitivity results of local and global nature, and we investigate how certain structural properties of the input signal are reflected in the corresponding feature vectors. Our theory applies to general filters and general Lipschitz-continuous non-linearities and pooling operators. Experiments on handwritten digit classification and facial landmark detection---including feature importance evaluation---complement the theoretical findings.

Finding a fixed point to a nonexpansive operator, i.e., $x^*=Tx^*$, abstracts many problems in numerical linear algebra, optimization, and other areas of scientific computing. To solve fixed-point problems, we propose ARock, an algorithmic framework in which multiple agents (machines, processors, or cores) update $x$ in an asynchronous parallel fashion. Asynchrony is crucial to parallel computing since it reduces synchronization wait, relaxes communication bottleneck, and thus speeds up computing significantly. At each step of ARock, an agent updates a randomly selected coordinate $x_i$ based on possibly out-of-date information on $x$. The agents share $x$ through either global memory or communication. If writing $x_i$ is atomic, the agents can read and write $x$ without memory locks. Theoretically, we show that if the nonexpansive operator $T$ has a fixed point, then with probability one, ARock generates a sequence that converges to a fixed points of $T$. Our conditions on $T$ and step sizes are weaker than comparable work. Linear convergence is also obtained. We propose special cases of ARock for linear systems, convex optimization, machine learning, as well as distributed and decentralized consensus problems. Numerical experiments of solving sparse logistic regression problems are presented.

Sparse matrix factorization is a popular tool to obtain interpretable data decompositions, which are also effective to perform data completion or denoising. Its applicability to large datasets has been addressed with online and randomized methods, that reduce the complexity in one of the matrix dimension, but not in both of them. In this paper, we tackle very large matrices in both dimensions. We propose a new factorization method that scales gracefully to terabyte-scale datasets. Those could not be processed by previous algorithms in a reasonable amount of time. We demonstrate the efficiency of our approach on massive functional Magnetic Resonance Imaging (fMRI) data, and on matrix completion problems for recommender systems, where we obtain significant speedups compared to state-of-the art coordinate descent methods. Matrix factorization is a flexible tool for uncovering latent factors in low-rank or sparse models. For instance, building on low-rank structure, it has proven very powerful for matrix completion, e.g. in recommender systems (Srebro et al., 2004; Candès & Recht, 2009). In signal processing and computer vision, matrix factorization with a sparse regularization is often called dictionary learning and has proven very effective for denoising and visual feature encoding (see Mairal, 2014, for a review).

We study the problem of off-policy value evaluation in reinforcement learning (RL), where one aims to estimate the value of a new policy based on data collected by a different policy. This problem is often a critical step when applying RL to real-world problems. Despite its importance, existing general methods either have uncontrolled bias or suffer high variance. In this work, we extend the doubly robust estimator for bandits to sequential decision-making problems, which gets the best of both worlds: it is guaranteed to be unbiased and can have a much lower variance than the popular importance sampling estimators. We demonstrate the estimator's accuracy in several benchmark problems, and illustrate its use as a subroutine in safe policy improvement. We also provide theoretical results on the inherent hardness of the problem, and show that our estimator can match the lower bound in certain scenarios.

In his book The Singularity Is Near, American computer scientist Ray Kurzweil had predicted a decade ago that by 2045 non-biological intelligence will have exceeded biological intelligence on Earth due to exponential changes in infotech, biotech and nanotech. Basically, man and machine will become one. But Murray Shanahan, a London-based cognitive robotics professor, disagrees with Kurzweil's theory in his more recent book, Technological Singularity. "Kurzweil was very precise (about time)," Shanahan tells China Daily in an interview in Beijing. "Technological singularity has a very dramatic impact on humanity."

The recent media frenzy about artificial intelligence (AI) has been unavoidable. This vision has perhaps come a step closer with the arrival of IBM Watsoni and Richard Susskind's latest book, The Future of the Professionsii, which predicts an internet society with greater virtual interaction with professional services such as doctors, teachers, accountants, architects and lawyers. In reality, is AI many years away from making any real impact in the legal sector? And should law firms see this technical advancement as an opportunity or threat? Broadly speaking, AI is the theory and development of computer systems which will perform tasks that normally require human intelligence.

The wildfire in Alberta, Canada, seems to be diminishing and residents should be able to return to the city of Fort McMurray over the next two weeks. The fire had appeared to be out of control just a few days ago but thanks to favourable weather conditions appears under control. The weather has played a huge part, but what about technology? AI, drones and satellites have all been used. Dr Guillermo Rein, from Imperial College, London and Editor-in-Chief of the journal Fire Technology explains how tech is now incorporated in fire management.

The sun is a powerful source of energy. So powerful, in fact, that Damon McMillan is betting his boat can cross more than 2,000 miles of ocean using only its rays. McMillan is the captain of Seacharger, which is on a mission to become the first unmanned, autonomous boat to cross an ocean using only solar power. It's a project that McMillan and three of his friends have worked on for two and a half years, after being inspired by a robotic sailboat competition. Constructing the boat took a lot of trial and error, McMillan says, and at times it seemed an impossible task.

Debugging code is a perennial headache for software developers, but scientists have announced a new technique that could make the process significantly easier. Developed at MIT's Computer Science and Artificial Intelligence Laboratory and the University of Maryland, the method essentially bridges the gap between the traditional technique of symbolic execution and today's modern software, making it possible to debug code far more efficiently. Symbolic execution is a software-analysis technique that can be used to locate and repair bugs automatically by tracing out every path a program might take during execution. The problem is, that technique doesn't tend to work well with applications written using today's programming frameworks. That's because modern applications generally import functions from those frameworks, which include huge libraries of frequently reused code.

Although Foxconn confirmed to the BBC that it was working to automate much of its manufacturing operations, the company denied that the new robotic assembly line would mean fewer jobs for humans. Instead, the company says it is simply using the machines to "replace repetitive tasks previously done by employees" while allowing those employees to focus on more valuable parts of the manufacturing process like R&D and quality control. "We will continue to harness automation and manpower in our manufacturing operations," Foxconn told the BBC, "and we expect to maintain our significant workforce in China." Meanwhile, the South China Morning Post also reports that 35 Taiwanese companies including Foxconn have spent a total of 4 billion yuan (or about 609 million USD) on artificial intelligence last year. Many of those companies employ tens of thousands in Kunshan, where two-thirds of the 2.5 million people are migrant workers. According to a government survey, 600 companies in Kunshan plan to follow Foxconn's lead.