Most recent improvements in NLP come from changes to the neural network architectures modeling the text input.Yet, state-of-theart models often rely on simple approaches to model the label space, e.g.bigram Conditional Random Fields (CRFs) in sequence tagging.More expressive graphical models are rarely used due to their prohibitive computational cost.In this work, we present an approach for efficiently training and decoding hybrids of graphical models and neural networks based on Gibbs sampling.Our approach is the natural adaptation of SampleRank (Wick et al., 2011) to neural models, and is widely applicable to tasks beyond sequence tagging.We apply our approach to named entity recognition and present a neural skipchain CRF model, for which exact inference is impractical.The skip-chain model improves over a strong baseline on three languages from CoNLL-02/03.We obtain new state-of-the-art results on Dutch. 1

Medical devices equipped with wireless connectivity and remote monitoring features are increasingly becoming connected to each other, to an outside programmer and even to the Internet. While Internet of Things technology enables health-care professionals to fine tune or modify medical device settings without invasive procedures, this also opens up large attack surfaces and introduces potential security vulnerabilities. Medical device hacks are slowly becoming a reality and it becomes more critical than ever to defend and protect these devices from security attacks. In this paper, we assess the feasibility of using machine learning models to efficiently determine attacks targeted on a medical device. Specifically, we develop feature sets to accurately profile a medical device and observe any deviation from its normal behavior. We test our method using different machine learning algorithms and provide a comparison analysis of the detection results.

Social influence has been a widely accepted phenomenon in social networks for decades. In this paper, we study influence from the perspective of structure, and focus on the simplest group structure - triad. We analyze two different genres of behavior: Retweeting on Weibo and Paying on CrossFire. We have several intriguing observations from these two networks. First, different internal structures of one's friends exhibit significant heterogeneity in influence patterns. Second, the strength of social relationship plays an important role in influencing one's behavior, and more interestingly, it is not necessarily positively correlated with the strength of social influence. We incorporate the triadic influence patterns into a predictive model to predict user's behavior. Experiment results show that our method can significantly improved the prediction accuracy.

Ten-finger freehand mid-air typing is a potential solution for post-desktop interaction. However, the absence of tactile feedback as well as the inability to accurately distinguish tapping finger or target keys exists as the major challenge for mid-air typing. In this paper, we present ATK, a novel interaction technique that enables freehand ten-finger typing in the air based on 3D hand tracking data. Our hypothesis is that expert typists are able to transfer their typing ability from physical keyboards to mid-air typing. We followed an iterative approach in designing ATK. We first empirically investigated users' mid-air typing behavior, and examined fingertip kinematics during tapping, correlated movement among fingers and 3D distribution of tapping endpoints. Based on the findings, we proposed a probabilistic tap detection algorithm, and augmented Goodman's input correction model to account for the ambiguity in distinguishing tapping finger. We finally evaluated the performance of ATK with a 4-block study. Participants typed 23.0 WPM with an uncorrected word-level error rate of 0.3% in the first block, and later achieved 29.2 WPM in the last block without sacrificing accuracy.