All networks, whether transport, telecommunications, IT hubs and even entire cities, need proper monitoring and management. Sophisticated processes (and the people who understand them) are required for the kind of deep network drill-down that reveals the causes of problems and ways in which they could be prevented. Similar processes are essential to handle any issues that do slip through the net, to analyze their effects, and to implement damage limitation and permanent solutions. Byte-level analysis of high-quality data makes these processes possible. For telecommunications network operators, such data are today generated in massive volumes.

I hate it when I hear senior executives state that they want to become data-driven, as if somehow having data is value in of itself. Now, one can hardly blame the unenlightened executive whose only perspectives on data are associated with statements like "Data is really the new oil" (Wall Street Journal) or "The world's most valuable resource is no longer oil, but data" (The Economist). The infatuation with "data-driven" versus "value-driven" can be confirmed from Google Trends (Figure 1). However, this is where the value determination of data and oil diverge. Oil has value as determined by General Acceptable Accounting Principles (GAAP).

A new analysis of Twitter and news coverage surrounding the Democratic primary candidates for the U.S. 2020 presidential elections shows that female candidates are attacked significantly more often than male candidates by trolls and fake news accounts. The report, published Nov. 5 by Lucina Di Meco, Global Fellow at The Wilson Center, used artificial intelligence in partnership with non-partisan data analytics firm Marvelous AI to track the coverage of six Democratic candidates on Twitter, measuring the volume of conversation around each candidate between December 2018 and April 2019. Joe Biden, Bernie Sanders, Pete Buttigieg, Elizabeth Warren, Kamala Harris and Amy Klobuchar were the candidates included in the study, which forms part of the broader report titled #ShePersisted: Women, Politics and Power in the New Media World. These online conversations were analyzed for one week after each candidate's official campaign launch between December 2018 and April 2019, depending on the candidate. Marvelous AI also examined the political bias and credibility of Twitter users participating in the conversation, as well as the themes and narratives surrounding each candidate.

In the age of artificial intelligence, predicting which jobs will fall to automation is as much about what machines can do as it is about what they can't. More than half of all jobs in America -- both blue and white-collar -- are resistant to automation, according to an acclaimed study published in 2013 by two Oxford University researchers. Co-author Carl Benedikt Frey, who directs Oxford's Technology and Employment program, broke down three areas where human intelligence still beats artificial intelligence: perception and manipulation, social intelligence; and creativity. Each type has what Frey calls a "bottleneck," which slows the pace at which certain workforces can be automated. The premise is simple: Technology won't replace human workers if it can't do the job.

Have you ever looked in your rear-view mirror and watched anxiously as a police car came up behind you? I'd dare say that most of us dread such a moment. It does not necessarily mean that you are a criminal or have done anything wrong. It's the notion that the police officer can potentially pull you over, referred to as a traffic stop, which gets us nervous and on-edge. Am I doing anything wrong in my driving, you right away begin to ponder. Is there anything about my car that might spark a traffic stop, you contemplate as your mind races trying to ascertain whether you are going to get pulled over or not. If the police car opts to go around you, it usually brings you a sense of momentary relief. Thank goodness, avoided getting stopped. For some drivers, once they realize that a police car is directly behind them, they will opt to switch lanes in hopes that the police car will merely go alongside and no longer sit behind their car. I know a few drivers that the minute they spot a police car even many cars behind them, they will right away try to maneuver into a lane that will keep them from perchance having the cops directly on their tail. Why do police perform these ad hoc traffic stops? In theory, the traffic stop is intended to ensure the safety of the roadways. If you are driving in a dangerous fashion, it seems sensible that having you pulled to the side of the road might prevent you from ramming into another car or running over a pedestrian. If your car is exhibiting some adverse condition and not fully safely drivable, suppose your exhaust pipe is hanging onto the ground and dragging along, this can create a traffic hazard for you and for other cars nearby. Probably handy to have a traffic stop to inform you about the matter and make sure that you are aware of it and take care of it.

This paper investigates the convergence of learning dynamics in Stackelberg games. In the class of games we consider, there is a hierarchical game being played between a leader and a follower with continuous action spaces. We establish a number of connections between the Nash and Stackelberg equilibrium concepts and characterize conditions under which attracting critical points of simultaneous gradient descent are Stackelberg equilibria in zero-sum games. Moreover, we show that the only stable critical points of the Stackelberg gradient dynamics are Stackelberg equilibria in zero-sum games. Using this insight, we develop a gradient-based update for the leader while the follower employs a best response strategy for which each stable critical point is guaranteed to be a Stackelberg equilibrium in zero-sum games. As a result, the learning rule provably converges to a Stackelberg equilibria given an initialization in the region of attraction of a stable critical point. We then consider a follower employing a gradient-play update rule instead of a best response strategy and propose a two-timescale algorithm with similar asymptotic convergence guarantees. For this algorithm, we also provide finite-time high probability bounds for local convergence to a neighborhood of a stable Stackelberg equilibrium in general-sum games. Finally, we present extensive numerical results that validate our theory, provide insights into the optimization landscape of generative adversarial networks, and demonstrate that the learning dynamics we propose can effectively train generative adversarial networks.

Many problems in statistical learning, imaging, and computer vision involve the optimization of a non-convex objective function with singularities at the boundary of the feasible set. For such challenging instances, we develop a new interior-point technique building on the Hessian-barrier algorithm recently introduced in Bomze, Mertikopoulos, Schachinger and Staudigl, [SIAM J. Opt. 2019 29(3), pp. 2100-2127], where the Riemannian metric is induced by a generalized self-concordant function. This class of functions is sufficiently general to include most of the commonly used barrier functions in the literature of interior point methods. We prove global convergence to an approximate stationary point of the method, and in cases where the feasible set admits an easily computable self-concordant barrier, we verify worst-case optimal iteration complexity of the method. Applications in non-convex statistical estimation and $L^{p}$-minimization are discussed to given the efficiency of the method.

This paper investigates to what extent we can improve reinforcement learning algorithms. Our study is split in three parts. First, our analysis shows that the classical asymptotic convergence rate $O(1/\sqrt{N})$ is pessimistic and can be replaced by $O((\log(N)/N)^{\beta})$ with $\frac{1}{2}\leq \beta \leq 1$ and $N$ the number of iterations. Second, we propose a dynamic optimal policy for the choice of the learning rate $(\gamma_k)_{k\geq 0}$ used in stochastic algorithms. We decompose our policy into two interacting levels: the inner and the outer level. In the inner level, we present the PASS algorithm (for "PAst Sign Search") which, based on a predefined sequence $(\gamma^o_k)_{k\geq 0}$, constructs a new sequence $(\gamma^i_k)_{k\geq 0}$ whose error decreases faster. In the outer level, we propose an optimal methodology for the selection of the predefined sequence $(\gamma^o_k)_{k\geq 0}$. Third, we show empirically that our selection methodology of the learning rate outperforms significantly standard algorithms used in reinforcement learning (RL) in the three following applications: the estimation of a drift, the optimal placement of limit orders and the optimal execution of large number of shares.

We define and study the problem of modular concept learning, that is, learning a concept that is a cross product of component concepts. If an element's membership in a concept depends solely on it's membership in the components, learning the concept as a whole can be reduced to learning the components. We analyze this problem with respect to different types of oracle interfaces, defining different sets of queries. If a given oracle interface cannot answer questions about the components, learning can be difficult, even when the components are easy to learn with the same type of oracle queries. While learning from superset queries is easy, learning from membership, equivalence, or subset queries is harder. However, we show that these problems become tractable when oracles are given a positive example and are allowed to ask membership queries. Keywords: Inductive Synthesis, Query-Based Learning, Modularity 1 Introduction Inductive synthesis or inductive learning is the synthesis of programs (concepts) from examples or other observations. Inductive synthesis has found application in formal methods, program analysis, software engineering, and related areas, for problems such as invariant generation (e.g.

Eye tracking is handled as one of the key technologies for applications which assess and evaluate human attention, behavior and biometrics, especially using gaze, pupillary and blink behaviors. One of the main challenges with regard to the social acceptance of eye-tracking technology is however the preserving of sensitive and personal information. To tackle this challenge, we employed a privacy-preserving framework based on randomized encoding to train a Support Vector Regression model on synthetic eye images privately to estimate human gaze. During the computation, none of the parties learns about the data or the result that any other party has. Furthermore, the party that trains the model cannot reconstruct pupil, blink or visual scanpath. The experimental results showed that our privacy preserving framework is also capable of working in real-time, as accurate as a non-private version of it and could be extended to other eye-tracking related problems.