In this paper we present a performance evaluation of machine learning and big data technologies in edge computing. Existing research focuses on deep learning methods on high-end edge computing devices. There have been few reports on classical machine learning methods on common edge computing devices. To close this gap, we evaluate support vector machine (SVM) and random forest (RF) on a Raspberry Pi platform. Our evaluation includes method execution time and data storage time. The latter is considered an important component in machine learning execution lifecycle and has not been jointly evaluated in existing research.

Recent studies have investigated how to achieve robustness for unsupervised domain adaptation (UDA). While most efforts focus on adversarial robustness, i.e. how the model performs against unseen malicious adversarial perturbations, robustness against benign common corruption (RaCC) surprisingly remains under-explored for UDA. Towards improving RaCC for UDA methods in an unsupervised manner, we propose a novel Distributionally and Discretely Adversarial Regularization (DDAR) framework in this paper. Formulated as a min-max optimization with a distribution distance, DDAR1is theoretically well-founded to ensure generalization over unknown common corruptions. Meanwhile, we show that our regularization scheme effectively reduces a surrogate of RaCC, i.e., the perceptual distance between natural data and common corruption. To enable a abetter adversarial regularization, the design of the optimization pipeline relies on an image discretization scheme that can transform "out-of-distribution" adversarial data into "in-distribution" data augmentation. Through extensive experiments, in terms of RaCC, our method is superior to conventional unsupervised regularization mechanisms, widely improves the robustness of existing UDA methods, and achieves state-of-the-art performance.

Distance based positioning methods have been widely used in today’s wireless networks for positioning network users. In this paper, we present a study on distance based positioning at network edges. We show that existing methods may not be able to find the optimal position at network edges due to the presence of measurement noise and the use of biased estimation. To handle this problem, we propose an improvement on the estimation method. Simulation results show that the proposed improvement can reduce position error by 30% in 20% of a network area.

Wi-Fi deployed inside a building can be used for positioning indoor users. A commonly used technology is weighted K-nearest neighbor (WKNN) fingerprint which positions a user based on K nearest reference points measured beforehand. The challenge lies in how to configure the value of K to obtain the best positioning accuracy. In this paper, we propose a self-adaptive WKNN (SAWKNN) algorithm with a dynamic K. By adjusting the value of K based on the signal strength, SAWKNN can obtain a better positioning accuracy than traditional WKNN. In particular, a significant percentage of the SAWKNN positioning makes use of a value K = 1. The performance of the proposed algorithm has been evaluated in real-world experiments.

Location spoofing and non-line-of-sight (NLOS) propagation are two leading reasons of serious localization errors in wireless networks. Previous studies have managed to identify these two factors separately. However, when present in the same system, these two factors can cause localization errors in a similar manner, making the identification difficult. In this paper, we address the problem of identifying location spoofing in NLOS conditions. We first carry out a geometric analysis on NLOS and derive a bound that can be used to differentiate NLOS from location spoofing. Based on the bound, we propose an identification method. We show that the proposed method is secure against different types of spoofing attacks including those from individuals and from multiple collaborative attackers. In particular, it can be used to identify the well-known “perfect location spoofing.” Simulation in wireless sensor networks indicates that our method can achieve a high accuracy with 0 false positive on identifying individual attacks and perfect location spoofing in NLOS conditions.

Various types of sensors have been embedded in smartphones such that a mobile user can easily conduct some sensing tasks. The mobile users conducting the sensing task with their sensor-equipped smartphones have their own unique features, thus can be efficiently complementary to stationary sensors which are deployed at specific locations. In this paper, we consider a heterogeneous sensor network composed of stationary sensors and mobile sensors (i.e., mobile users with sensor-equipped smartphones), in which a key question is how the service provider selects the sensors to conduct the sensing task considering the heterogeneity of sensors in terms of location, mobility pattern, energy constraint, and sensing cost. We propose a greedy algorithm GSSA to reduce the computational complexity. Simulation results show the nice performance of the proposed algorithms compared with the optimal sensor selection algorithm using dynamic programming. In specific, the proposed GSSA improves the achieved social welfare by 32.3% and 35.6% with the time period T=20 for high mobility and low mobility patterns, respectively, compared with the random selection.

Bad channel conditions can cause serious error in wireless localization. To identify this problem, current methods commonly make use of a consistency analysis. A similar consistency analysis has been adopted in recent studies of location spoofing to identify malicious users. Since they both utilize a consistency analysis, the following problem arise naturally -- how to identify location spoofing in bad channel conditions? In this paper, we present an analysis on the degree of inconsistency. We show that location spoofing will be associated with an inconsistency different from the ones caused by bad channel conditions. Based on this property, we propose an identification method to location spoofing in bad channel conditions. Simulation in wireless sensor networks shows that the proposed method can achieve an average identification rate of 98% with 0% false alarm.

Abstract There is recently an increasing interest in applications based on localization of mobile objects in outdoor environment. Many existing localization solutions rely primarily on a hybrid wireless network/dead reckoning (DR) scheme, as the regular wireless network can hardly estimate a position with satisfactory accuracy in bad channel conditions. However, the DR can involve considerable hardware investments and operating costs, moreover, it suffers from serious error accumulations in motion measurements of the object. To remedy drawbacks of the hybrid scheme, we present a mobility enhanced localization (MEL), in which the mobile object serves as a pseudo beacon. The proposed scheme enables network localization in those areas where traditional wireless networks may not work, and thus, avoids large investments the special hardware cost and error accumulations of DR. We further propose a weighted preprocessing method for classical localization algorithms, since theoretical analysis shows that they fail to consider some boundary conditions. Extensive real world Global Positioning System (GPS) experimental results demonstrate the superiority of the proposed MEL scheme. Copyright © 2008 John Wiley & Sons, Ltd.

One of the important issues in wireless localization is concerned with the identification of non-line-of-sight (NLOS) radio signal propagation, since NLOS propagation is considered the dominant source of localization error. The existing identification methods focus on "consistency" tests, based on the assumption that localization involving the use of NLOS radio signals cannot be performed in a consistent manner. However, the validity of the foregoing assumption has not been properly investigated. This paper presents a theoretical analysis of the localization using NLOS radio signals, and shows that the assumption would no longer be valid when the mobile user is located outside the convex hull of the underlying beacons. As a result, existing NLOS identification methods, as well as many localization approaches, could perform unsatisfactorily. The importance of the proposed convex hull condition for NLOS identification is confirmed by extensive simulation results.

Lately there exist increasing demands for online abnormality monitoring over trajectory streams, which are obtained from moving object tracking devices. This problem is challenging due to the requirement of high speed data processing within limited space cost. In this paper, we present a novel framework for monitoring anomalies over continuous trajectory streams. First, we illustrate the importance of distance-based anomaly monitoring over moving object trajectories. Then, we utilize the local continuity characteristics of trajectories to build local clusters upon trajectory streams and monitor anomalies via efficient pruning strategies. Finally, we propose a piecewise metric index structure to reschedule the joining order of local clusters to further reduce the time cost. Our extensive experiments demonstrate the effectiveness and efficiency of our methods.

Due to the advent of many location based applications, mobile localization in outdoor environments has recently attracted much attention. Many existing localization solutions rely primarily on a hybrid wireless network/dead reckoning (DR) method, as the regular wireless network can hardly estimate a location with satisfactory accuracy in bad channel conditions. However, the DR scheme can involve considerable hardware investments, moreover, it suffers from serious error accumulations in motion measurements of a mobile object. To remedy drawbacks of the hybrid method, we present a novel mobile localization scheme based on the concept of pseudo beacon. The proposed scheme can perform network localization in those areas where traditional wireless networks may not work, and thus, avoids large investments the special hardware cost and error accumulations of DR. Extensive real world GPS experimental results demonstrate the significant superiority of the proposed scheme.