University of Virginia

About: University of Virginia is a education organization based out in Charlottesville, Virginia, United States. It is known for research contribution in the topics: Population & Poison control. The organization has 52543 authors who have published 113268 publications receiving 5220506 citations. The organization is also known as: U of V & UVa.

Robust Distributed Spectrum Sensing in Cognitive Radio Networks

Abstract: Distributed spectrum sensing (DSS) enables a Cognitive Radio (CR) network to reliably detect licensed users and avoid causing interference to licensed communications. The data fusion technique is a key component of DSS. We discuss the Byzantine failure problem in the context of data fusion, which may be caused by either malfunctioning sensing terminals or Spectrum Sensing Data Falsification (SSDF) attacks. In either case, incorrect spectrum sensing data will be reported to a data collector which can lead to the distortion of data fusion outputs. We investigate various data fusion techniques, focusing on their robustness against Byzantine failures. In contrast to existing data fusion techniques that use a fixed number of samples, we propose a new technique that uses a variable number of samples. The proposed technique, which we call Weighted Sequential Probability Ratio Test (WSPRT), introduces a reputation-based mechanism to the Sequential Probability Ratio Test (SPRT). We evaluate WSPRT by comparing it with a variety of data fusion techniques under various network operating conditions. Our simulation results indicate that WSPRT is the most robust against the Byzantine failure problem among the data fusion techniques that were considered.

The ACS Survey of Galactic Globular Clusters. I. Overview and Clusters without Previous Hubble Space Telescope Photometry

Abstract: We present the first results of a large Advanced Camera for Surveys (ACS) survey of Galactic globular clusters. This Hubble Space Telescope (HST) Treasury project is designed to obtain photometry with S/N (signal-to-noise ratio) 10 for main-sequence stars with masses 0.2 M⊙ in a sample of globulars using the ACS Wide Field Channel. Here we focus on clusters without previous HST imaging data. These include NGC 5466, NGC 6779, NGC 5053, NGC 6144, Palomar 2, E3, Lynga 7, Palomar 1, and NGC 6366. Our color-magnitude diagrams (CMDs) extend reliably from the horizontal branch to as much as 7 mag fainter than the main-sequence turnoff and represent the deepest CMDs published to date for these clusters. Using fiducial sequences for three standard clusters (M92, NGC 6752, and 47 Tuc) with well-known metallicities and distances, we perform main-sequence fitting on the target clusters in order to obtain estimates of their distances and reddenings. These comparisons, along with fitting the cluster main sequences to theoretical isochrones, yield ages for the target clusters. We find that the majority of the clusters have ages that are consistent with the standard clusters at their metallicities. The exceptions are E3, which appears ~2 Gyr younger than 47 Tuc, and Pal 1, which could be as much as 8 Gyr younger than 47 Tuc.

Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints

Abstract: Language is increasingly being used to de-fine rich visual recognition problems with supporting image collections sourced from the web. Structured prediction models are used in these tasks to take advantage of correlations between co-occurring labels and visual input but risk inadvertently encoding social biases found in web corpora. In this work, we study data and models associated with multilabel object classification and visual semantic role labeling. We find that (a) datasets for these tasks contain significant gender bias and (b) models trained on these datasets further amplify existing bias. For example, the activity cooking is over 33% more likely to involve females than males in a training set, and a trained model further amplifies the disparity to 68% at test time. We propose to inject corpus-level constraints for calibrating existing structured prediction models and design an algorithm based on Lagrangian relaxation for collective inference. Our method results in almost no performance loss for the underlying recognition task but decreases the magnitude of bias amplification by 47.5% and 40.5% for multilabel classification and visual semantic role labeling, respectively。