A BSTRACT Off-policy stochastic actor-critic methods rely on approximating the stochastic policy gradient in order to derive an optimal policy. One may also derive the optimal policy by approximating the action-value gradient. The use of action-value gradients is desirable as policy improvement occurs along the direction of steepest ascent. This has been studied extensively within the context of natural gradient actor-critic algorithms and more recently within the context of deterministic policy gradients. In this paper we briefly discuss the off-policy stochastic counterpart to deterministic action-value gradients, as well as an incremental approach for following the policy gradient in lieu of the natural gradient.

Deep neural network algorithms are difficult to analyze because they lack structure allowing to understand the properties of underlying transforms and invariants. Multiscale hierarchical convolutional networks are structured deep convolutional networks where layers are indexed by progressively higher dimensional attributes, which are learned from training data. Each new layer is computed with multidimensional convolutions along spatial and attribute variables. We introduce an efficient implementation of such networks where the dimensionality is progressively reduced by averaging intermediate layers along attribute indices. Hierarchical networks are tested on CIFAR image data bases where they obtain comparable precisions to state of the art networks, with much fewer parameters. We study some properties of the attributes learned from these databases.

In this paper, we investigate the common scenario where every candidate item for recommendation is characterized by a maximum capacity, i.e., number of seats in a Point-of-Interest (POI) or size of an item's inventory. Despite the prevalence of the task of recommending items under capacity constraints in a variety of settings, to the best of our knowledge, none of the known recommender methods is designed to respect capacity constraints. To close this gap, we extend three state-of-the art latent factor recommendation approaches: probabilistic matrix factorization (PMF), geographical matrix factorization (GeoMF), and bayesian personalized ranking (BPR), to optimize for both recommendation accuracy and expected item usage that respects the capacity constraints. We introduce the useful concepts of user propensity to listen and item capacity. Our experimental results in real-world datasets, both for the domain of item recommendation and POI recommendation, highlight the benefit of our method for the setting of recommendation under capacity constraints.

Many modern databases include personal and sensitive correlated data, such as private information on users connected together in a social network, and measurements of physical activity of single subjects across time. However, differential privacy, the current gold standard in data privacy, does not adequately address privacy issues in this kind of data. This work looks at a recent generalization of differential privacy, called Pufferfish, that can be used to address privacy in correlated data. The main challenge in applying Pufferfish is a lack of suitable mechanisms. We provide the first mechanism -- the Wasserstein Mechanism -- which applies to any general Pufferfish framework. Since this mechanism may be computationally inefficient, we provide an additional mechanism that applies to some practical cases such as physical activity measurements across time, and is computationally efficient. Our experimental evaluations indicate that this mechanism provides privacy and utility for synthetic as well as real data in two separate domains.

How spiking networks are able to perform probabilistic inference is an intriguing question, not only for understanding information processing in the brain, but also for transferring these computational principles to neuromorphic silicon circuits. A number of computationally powerful spiking network models have been proposed, but most of them have only been tested, under ideal conditions, in software simulations. Any implementation in an analog, physical system, be it in vivo or in silico, will generally lead to distorted dynamics due to the physical properties of the underlying substrate. In this paper, we discuss several such distortive effects that are difficult or impossible to remove by classical calibration routines or parameter training. We then argue that hierarchical networks of leaky integrate-and-fire neurons can offer the required robustness for physical implementation and demonstrate this with both software simulations and emulation on an accelerated analog neuromorphic device.

Inferring macroscopic properties of physical systems from their microscopic description is an ongoing work in many disciplines of physics, like condensed matter, ultra cold atoms or quantum chromo dynamics. The most drastic changes in the macroscopic properties of a physical system occur at phase transitions, which often involve a symmetry breaking process. The theory of such phase transitions was formulated by Landau as a phenomenological model [1] and later devised from microscopic principles using the renormalization group [2, 3]. One can identify phases by knowledge of an order parameter which is zero in the disordered phase and nonzero in the ordered phase. Whereas in many known models the order parameter can be determined by symmetry considerations of the underlying Hamiltonian, there are states of matter where such a parameter can only be defined in a complicated non-local way [4]. These systems include topological states like topological insulators, quantum spin hall states [5] or quantum spin liquids [6].

First of all, I will explain the reason for the post title. For those who have not seen the films, I summarize: "A group of four illusionists win year after year to the public with their incredible magic shows and even mocking the FBI. GSMA is a great illusionist and MWC is their principal magic show. We are invited year after year to visit an event with unique keynote speakers, an enormous list of exhibitors, amazing performances and a great LinkedInplace where we can meet in person some of our social media contacts. What else can we ask for? I know that it is very ruthless to compare the GSMA with illusionists and the MWC as their greatest magic show, but at least I see quite a few reasonable resemblances, you don t. If in 2015 I wrote " MWC 2015: Everything Connected, Tapas and Jamon", and I argued as one the reasons to attend MWC was the fact it was celebrated in Barcelona. In 2016, in my post "GSMA need to think how to reinvent MWC" I justify the reasons why the MWC needed to reinvent itself. One thing has become clear to me after many years attending MWCs, this is the world's biggest phone and mobile networks show, with manufacturers set to unveil a raft of new phone handsets and new technology. However, the GSMA had insisted on introducing more and more distractions like Internet of Things (IoT), Connected Living, Connected Car, AR/ VR, Robots. Maybe the reason is because Telecom operators do not have the DNA to change. Still, many telecom operators take a dim view of some of the aggressive moves being made by these peers, especially when it comes to business models based on commercializing customer data. "I expected to see less hype and a dose of common sense" Starting by the announcement of Spain's Telefonica to introduce a broad plan "4th Platform" to help both consumer and business customers keep greater control over their data rather than giving it away to web giants Google, Facebook and Amazon. "I expected to see more applications where IoT will become a lot less exciting, but more useful and profitable.

Another day, another fun internet thing that uses neural networks for facial manipulation. This time it's DeepWarp, a demo created by Yaroslav Ganin, Daniil Kononenko, Diana Sungatullina, and Victor Lempitsky, that uses deep architecture to move human eyeballs in a still image. First spotted by Prosthetic Knowledge, DeepWarp is focused on realistic "gaze manipulation." The authors of the demo acknowledge that similar projects already exist (like the smile-manipulator FaceApp), but without such a singular, detailed focus. The authors note that their findings in this study could be applied to solve real-world issues of eye movement, like for "gaze correction in video conferencing."

Amazon's Alexa already boasts more than 10,000 skills, but has now added medical advice to its repertoire. WebMD announced today that it's launched its own skill for all Alexa-enabled devices (including the Echo, Echo Dot, and Fire TV), which can answer basic health-related queries. Topics include treatments for common ailments ("Alexa, ask WebMD how to treat a sore throat"), definitions of basic diseases ("Alexa, ask WebMD what diabetes is"), and the side effects of certain drugs ("Alexa, ask WebMD to tell me about amoxicillin"). WebMD stresses that, like its website, the new Alexa skill is only meant to offer supplementary information, and adds that the software is a work-in-progress. "We want to be in the place where we believe computing is going," WebMD's vice president of mobile products, Ben Greenberg, tells The Verge.

Watson captured the public imagination about artificial intelligence after defeating two world champions of Jeopardy in 2011 and bringing home a $1 million prize. Since then, Watson has gained new cognitive capabilities through APIs like Alchemy (for sentiment analysis), Tone Analyzer (for personality and emotional analysis), and Conversation (a chatbot builder) and has been embedded in hundreds of applications across financial services, healthcare, retail, and digital. "Chances are, you've interacted with Watson without realizing it," says Alyssa Simpson, Program Director at IBM Watson. "Many companies hide their use of the technology for competitive reasons. They don't want a competitor to buy it too."