While companies like Amazon pour considerable resources into finding ways of using drones to deliver such things as shoes and dog treats, Zipline has been saving lives in Rwanda since October 2016 with drones that deliver blood. Zipline's autonomous fixed-wing drones now form an integral part of Rwanda's medical-supply infrastructure, transporting blood products from a central distribution center to hospitals across the country. And in 2018, Zipline's East African operations will expand to include Tanzania, a much larger country. Delivering critical medical supplies in this region typically involves someone spending hours (or even days) driving a cooler full of life-saving medicine or blood along windy dirt roads. Such deliveries can become dangerous or even impossible to make if roads and bridges get washed out.

We consider an extension of the set covering problem (SCP) introducing (i)~multicover and (ii)~generalized upper bound (GUB)~constraints. For the conventional SCP, the pricing method has been introduced to reduce the size of instances, and several efficient heuristic algorithms based on such reduction techniques have been developed to solve large-scale instances. However, GUB constraints often make the pricing method less effective, because they often prevent solutions from containing highly evaluated variables together. To overcome this problem, we develop heuristic algorithms to reduce the size of instances, in which new evaluation schemes of variables are introduced taking account of GUB constraints. We also develop an efficient implementation of a 2-flip neighborhood local search algorithm that reduces the number of candidates in the neighborhood without sacrificing the solution quality. In order to guide the search to visit a wide variety of good solutions, we also introduce a path relinking method that generates new solutions by combining two or more solutions obtained so far. According to computational comparison on benchmark instances, the proposed method succeeds in selecting a small number of promising variables properly and performs quite effectively even for large-scale instances having hard GUB constraints.

Abstract--We consider the problem of tracking an intruder using a network of wireless sensors. For tracking the intruder at each instant, the optimal number and the right configuration of sensors has to be powered. As powering the sensors consumes energy, there is a trade off between accurately tracking the position of the intruder at each instant and the energy consumption of sensors. This problem has been formulated in the framework of Partially Observable Markov Decision Process (POMDP) [1]. Even for the simplest model considered in [1], the curse of dimensionality renders the problem intractable. We formulate this problem with a suitable state-action space in the framework of POMDP and develop a reinforcement learning algorithm utilizing the Upper Confidence Tree Search (UCT) method to mitigate the state-action space explosion. Through simulations, we illustrate that our algorithm yields good performance and scales well with the increasing state and action space. I. INTRODUCTION The problem of detecting an intruder (Intrusion Detection (ID) problem) using a network of sensors arises in various applications like tracking the movement of wild animals in the forest, house/shop surveillance for safety and security and so on. In this problem, the objective of the ID system is to track one or more intruders moving in the field of a wireless sensor network (WSN). Typically, WSNs operate on limited power supply.

Different from other sequential data, sentences in natural language are structured by linguistic grammars. Previous generative conversational models with chain-structured decoder ignore this structure in human language and might generate plausible responses with less satisfactory relevance and fluency. In this study, we aim to incorporate the results from linguistic analysis into the process of sentence generation for high-quality conversation generation. Specifically, we use a dependency parser to transform each response sentence into a dependency tree and construct a training corpus of sentence-tree pairs. A tree-structured decoder is developed to learn the mapping from a sentence to its tree, where different types of hidden states are used to depict the local dependencies from an internal tree node to its children. For training acceleration, we propose a tree canonicalization method, which transforms trees into equivalent ternary trees. Then, with a proposed tree-structured search method, the model is able to generate the most probable responses in the form of dependency trees, which are finally flattened into sequences as the system output. Experimental results demonstrate that the proposed X2Tree framework outperforms baseline methods over 11.15% increase of acceptance ratio.

Adversarial learning of probabilistic models has recently emerged as a promising alternative to maximum likelihood. Implicit models such as generative adversarial networks (GAN) often generate better samples compared to explicit models trained by maximum likelihood. Yet, GANs sidestep the characterization of an explicit density which makes quantitative evaluations challenging. To bridge this gap, we propose Flow-GANs, a generative adversarial network for which we can perform exact likelihood evaluation, thus supporting both adversarial and maximum likelihood training. When trained adversarially, Flow-GANs generate high-quality samples but attain extremely poor log-likelihood scores, inferior even to a mixture model memorizing the training data; the opposite is true when trained by maximum likelihood. Results on MNIST and CIFAR-10 demonstrate that hybrid training can attain high held-out likelihoods while retaining visual fidelity in the generated samples.

In Stackelberg security games, a defender seeks to randomly allocate limited security resources to protect critical targets from an attack. In this paper, we study a fundamental, yet underexplored, phenomenon in security games, which we term the \emph{Curse of Correlation} (CoC). Specifically, we observe that there are inevitable correlations among the protection status of different targets. Such correlation is a crucial concern, especially in \emph{spatio-temporal} domains like conservation area patrolling, where attackers can surveil patrollers at certain areas and then infer their patrolling routes using such correlations. To mitigate this issue, we propose to design entropy-maximizing defending strategies for spatio-temporal security games, which frequently suffer from CoC. We prove that the problem is \#P-hard in general. However, it admits efficient algorithms in well-motivated special settings. Our experiments show significant advantages of max-entropy algorithms over previous algorithms. A scalable implementation of our algorithm is currently under pre-deployment testing for integration into FAMS software to improve the scheduling of US federal air marshals.

If there's one thing we can learn from the Terminator franchise, it's that too much technological advancement is not something we should be a hundred percent on board with. What's worse is that with the advancement we're getting with the AI technology, the creation of Skynet can actually be not that far ahead into the future. We're slowly getting there, and that should terrify you. Not sold on the idea yet? In order to start the ball rolling, here are some terrifying facts about Artificial Intelligence right now.

Venture capitalists lavished $10.8 billion on artificial intelligence and machine learning technology companies in 2017, according to PitchBook Data Inc. They've placed major bets that AI innovation can't go far or fast enough to meet demand. But controversial use cases -- like when algorithms decide the fate of the criminally tried -- and the danger of coded-in bias suggest it's gone too far already without regulatory oversight. "This technology's coming at us so fast, we don't have all the policies figured out," said Beena Ammanath (pictured), global vice president of big data, artificial intelligence and new tech innovation at Hewlett Packard Enterprise Co. While consumer, business and government users embrace AI software that makes their jobs and lives simpler, they are simultaneously tasked with building the guardrails around the tech.

At the 2017 HR Technology Expo and Conference, Aberdeen witnessed something startling: Human Capital Management (HCM) technology vendors were downplaying AI as they described how they were catapulting their solution agenda further ahead into the 21st century. While one of the reasons for this is that the technology is not yet living up to the visions over which Wall Street is hopelessly salivating, the reality is that HCM vendors are not on board with the elimination of people from the workforce. It simply doesn't jibe with their goals and the way they see the market. While themes related to machine learning and AI came up in conversations with these technologists, HCM vendors seemed more excited about the way advanced-stage analytics and predictive capabilities rooted in trend analyses were helping people use technology more effectively. In other words, technologists are more interested in how their creations are finally improving labor productivity.

Artificial Intelligence and Machine Learning, or AI and ML for short respectively, are two terms which get thrown around quite often these days. Most people think that they share the same meaning, and while they are quite closely associated, one cannot be used instead of the other. Both of them crop up frequently when discussing analytics, data or any sort of technological change, so I think it's important we settle what they actually mean once and for all. In short, artificial intelligence is the general concept where machines are able to carry out tasks in a "smart" way. Or, at least, in a way which we would consider smart.