In this work, we introduce the idea of “mutually exclusive learning” and formulate it as a multi-label classification problem. We provide two implementations of the mutually exclusive learner; the first using a consensus based multilabel framework and the second using an adaptive multi-label framework. We use the mutually exclusive learning paradigm for designing both generators and discriminators (classifiers). We experiment with the MNIST dataset for both classifying the digits using a mutually exclusive classifier and generating handwritten digits using a generative adversarial network (GAN) trained with a mutually exclusive discriminator implementation. Our results establish that GANs trained with a mutually exclusive discriminator converge faster than the corresponding GAN with a standard discriminator. Furthermore, the quality of the generated images is also visually better than those generated by a regular GAN.

We propose a deep learning framework for solving the problem of position and orientation estimation (DeePOE) of a radio frequency (RF) transmitter using the in-phase$(I)$and quadrature-phase$(Q)$components of the RF signal data. Our goal is to demonstrate a proof of concept system with an end-to-end implementation in order to overcome the shortcomings of state-of-the-art joint position and orientation estimation systems. The proposed DeePOE framework consists of a convolutional neural network (CNN) which is designed to exploit latent features present within the received raw I/Q signal data. This enables receivers equipped with the DeePOE framework to predict the position and orientation of a transmitter, relative to itself in a predefined coordinate system, solely from physical layer information. DeePOE jointly optimizes the position and orientation estimation objectives using transfer learning, iteratively over the training epochs. In order to validate and refine the DeePOE framework, we perform real-world (indoor and outdoor) experiments using 16 GB of raw I/Q data collected with directional emitters placed in various orientations and at different distances (positions) from both directional and omnidirectional receivers. The framework achieves on average a F1score of 0.922 for the task of predicting 12 orientations from the data collected using an omnidirectional antenna. It also yields F1score of 0.847 for data collected with a directional antenna which involves predicting 48 orientations. DeePOE achieves on average F1score of 0.963 for predicting the transmitter position with respect to the receiver placed at a known location, for all the cases.

Network cascades represent a number of real-life applications: social influence, electrical grid failures, viral spread, and so on. The commonality between these phenomena is that they begin from a set of seed nodes and spread to other regions of the network. We consider a variant of a critical node detection problem dubbed the robust critical node fortification problem, wherein the decision maker wishes to fortify nodes (within a budget) to limit the spread of cascading behavior under uncertain conditions. In particular, the arc weights—how much influence one node has on another in the cascade process—are uncertain but are known to lie in some range bounded by a worst-case budget uncertainty. This problem is shown to be [Formula: see text]-hard even in the deterministic case. We formulate a mixed-integer program (MIP) to solve the deterministic problem and improve its continuous relaxation via nonlinear constraints and convexification. The robust problem is computationally more difficult, and we present an MIP-based expand-and-cut exact solution algorithm, in which the expansion is enhanced by cutting planes, which are themselves tied to the expansion process. Insights from these exact solutions motivate two novel (interrelated) centrality measures, and a centrality-based heuristic that obtains high-quality solutions within a few seconds. Finally, extensive computational results are given to validate our theoretical developments as well as provide insights into structural properties of the robust problem and its solution.

We introduce a robust neural network based method for classifying aircraft, using raw I/Q data obtained from the Automatic Dependent Surveillance-Broadcast (ADS-B) data from airplanes. ADS-B has become the de-facto standard for air traffic control and forms the basis of the Next Generation Air Transportation System (NextGen). Although ADS-B is at the core of modern day air traffic control, the standard lacks basic security features such as encryption and authentication. As a result, it is possible to spoof ADS-B data and in the process create unprecedented operational havoc in the skies. In this work we propose FlightSense: a robust adversarial learning based system for filtering out spoofed ADS-B data and subsequent identification of airplanes operating in the airspace from the filtered signal. We use the framework of a generative adversarial network (GAN) for our implementation, which is end-to-end in that it uses the raw I/Q signal data as input and no preprocessing steps are required. We present experiments and results to demonstrate the efficacy of our methods using a real world standardized ADS-B dataset.

Computation of accurate indoor positioning information is important in several areas like mobile robotics, large scale sensor networks, smart city, virtual reality and applications involving internet-of-things (IoT). In spite of its growing importance due to the advent of large scale autonomous deployments of sensor networks and IoTs, we still do not have a solution for indoor positioning that is equivalent to the Global Positioning System (GPS), which is the standard for large scale outdoor 10- calization. However GPS is not useful for indoor positioning as it is often unreliable and/or unavailable in indoor environments. In this paper we present a multi-source radio frequency (RF) based framework for automatic indoor positioning using received signal strength (RSS) and demonstrate its efficacy by simultaneously using broadcast FM radio & GSM signals for position estimation. Our framework is data driven and can be justified using Bayesian minimum risk analysis and is easy to extend by incorporating other sources of RF signals (like WiFi signals) and thus provides a generalized framework for building indoor positioning systems using signals of opportunity. We call our framework RF-MSiP: Radio Frequency Multi-Source Indoor Positioning Using our algorithms with the well known AMBILOC dataset, we can localize exactly for approximately 98.7% of the test locations over a period of one year, which demonstrates not only the efficacy of the algorithms but also its resiliency to change in the RF environment across the year, thus establishing the transfer learning capability of our system.

Primary user emulation (PUE) attacks can pose a significant threat to the deployment of a robust cognitive radio network implementing dynamic spectrum access, for an intelligent allocation and usage of already crowded spectrum bands. In this paper, we present a solution towards the PUE attacks. We present two generative adversarial net (GAN) based models to successfully emulate the primary users (PUs) in two ways. We propose a (i) dumb generator model without any "prior" knowledge of PU's feature space, (ii) a smart generator model with some "prior" knowledge about PU's transmission. We also propose two deep neural network based discriminator models to discriminate between the PU and the emulated primary users (EPU) from the corresponding generators. Both the generator and discriminator of each GAN model gets smarter with iterative and sequential GAN training. Through a testbed evaluation, we show that discriminators are able to catch ~50% of PUE attackers without the GAN training during the deployment phase. We also observe 100% accuracy for both the GAN models during training phase. Ultimately, after the GAN training, the discriminators achieved 98% and 99.5% accuracies, for dumb and smart generator models respectively, to distinguish "yet to be seen" PUE attacker.

The recent advances of wireless technologies in RF environments coupled with large scale usage of such technologies has warranted more autonomous deployments of wireless systems. Machine learning techniques, that include recurrent structures, have shown promise in creating such autonomous deployments using the idea of Radio Frequency Machine Learning (RFML). In large scale autonomous deployments of wireless communication networks, the signals received from one component play a crucial role in the decision making process of other components. In order to efficiently implement such systems each component of the network should be uniquely identifiable. In this paper we propose a transmitter fingerprinting technique for radio device identification using recurrent structures, by exploiting the temporal property of the received radio signal. We design and implement three recurrent neural networks (RNNs) using different types of cell models: (i) long short term memory (LSTM); (ii) gated recurrent unit (GRU) and (iii) convolutional long short term memory (ConvLSTM), for this task. We program 8 universal software radio peripheral (USRP) software defined radios (SDRs) as transmitters and collect over-the-air raw in-phase (I) and quadrature (Q) (I/Q) time series data from them using a DVB-T RTL-SDR receiver, in a laboratory setting. We exploit both the temporal variations as well as the inherent spatial dependencies in the collected I/Q time series data, to learn unique feature representations and use these as "fingerprints'" for identifying the transmitters. Experimental results reveal that the RNNs with LSTM, GRU, and ConvLSTM cells are able to correctly distinguish between the 8 transmitters with 92%, 95.3%, 97.2% accuracy respectively.

Understanding and analyzing the radio frequency (RF) environment have become indispensable for various autonomous wireless deployments. To this end, machine learning techniques have become popular as they can learn, analyze and even predict the RF signals and associated parameters that characterize a RF environment. However, classical machine learning methods have their limitations and there are situations where such methods become ineffective. One such setting is where active adversaries are present and try to disrupt the RF environment through malicious activities like jamming or spoofing. In this paper we propose an adversarial learning technique for identifying rogue RF transmitters and classifying trusted ones by designing and implementing generative adversarial nets (GAN). The GAN exploits the in-phase (I) and quadrature imbalance (i.e., the IQ imbalance) present in all transmitters to learn the unique high dimensional features that can be used as “fingerprints” for identifying and classifying the transmitters. We implement a generative model that learns the sample space of the IQ values of the known transmitters and use the learned representation to generate fake signals that imitate the transmissions of the known transmitters. We program 8 universal software radio peripheral (USRP) software defined radios as trusted transmitters and collect over-the-air raw IQ data from them using a RTL-SDR in a laboratory setting. We also implement a discriminator model and show that the discriminator is able to discriminate between the trusted transmitters from fake ones with 99.9% accuracy. Finally, the trusted transmitters are classified using convolutional neural network (CNN) and fully connected deep neural networks (DNN). Results reveal that the CNN and DNN are able to correctly discriminate between the 8 trusted transmitters with 81.6% and 96.6% accuracies respectively.

Closeness centrality is a class of distance-based measures in the network analysis literature to quantify reachability of a given vertex (or a group of vertices) by other network agents. In this paper, we consider a new class of critical edge detection problems, in which given a group of vertices that represent an important subset of network elements of interest (e.g., servers that provide an essential service to the network), the decision maker is interested in identifying a subset of critical edges whose removal maximally degrades the closeness centrality of those vertices. We develop a general optimization framework, in which the closeness centrality measure can be based on any nonincreasing function of distances between vertices, which, in turn, can be interpreted as communication efficiency between them. Our approach includes three well-known closeness centrality measures as special cases: harmonic centrality, decay centrality, and [Formula: see text]-step reach centrality. Furthermore, for quantifying the centrality of a group of vertices we consider three different approaches for measuring the reachability of the group from any vertex in the network: minimum distance to a vertex in the group, maximum distance to a vertex in the group, and the average centrality of vertices in the group. We study the theoretical computational complexity of the proposed models and describe the corresponding mixed integer programming formulations. For solving medium- and large-scale instances of the problem, we first develop an exact algorithm that exploits the fact that real-life networks often have rather small diameters. Then we propose two conceptually different heuristic algorithms. Finally, we conduct computational experiments with real-world and synthetic network instances under various settings, which reveal interesting insights and demonstrate the advantages and limitations of the proposed models and algorithms. The online appendices are available at https://doi.org/10.1287/ijoc.2018.0829 .

Abstract We consider the problem of detecting a collection of critical node structures of a graph whose deletion results in the maximum deterioration of the graph's connectivity. The proposed approach is aimed to generalize other existing models whose scope is restricted to removing individual and unrelated nodes. We consider two common metrics to quantify the connectivity of the residual graph: the total number of connected node pairs and the size of the largest connected component. We first discuss the computational complexity of the problem and then introduce a general mixed‐integer linear formulation, which depending on the kind of node structures, may have an exponentially large number of variables and constraints. To solve this potentially large model, we develop a branch‐price‐and‐cut framework, along with some valid inequalities and preprocessing algorithms to strengthen the formulation and reduce the overall execution time. We use the proposed approach to solve the problem for the cases, where the node structures form cliques or stars and provide further directions on how to extend the framework for detecting other kinds of critical structures as well. Finally, we test the quality of our approach by solving a collection of real‐life and randomly generated instances with various configurations, analyze the benefits of our model, and propose further enhancements.

In this paper we attempt to fuse both airborne Lidar and high resolution images (0.5 feet per pixel) to identify road networks in a large geographic region. We perform pixel-wise segmentation to classify each pixel as road or non-road based on color and depth features in a larger neighborhood context. This constitutes a bimodal setting because the RGB pixels represent the color space and the depth values come from three dimensional Lidar readings. We present multiple strategies for fusing Lidar and images. We describe a cost-effective, modular, deep convolution network design, TriSeg which gives better IoU metric for the aerial road segmentation problem than the state of the art RGB only architectures. We report on many other architectures as well as release our dataset for further research: https://bitbucket.org/biswas/fusion_lidar_images.

We present a distributed image based robot (re)-localization system with four non-stereo monocular cameras using a deep Convolutional Network (Convnet). Our system trains the well known Posenet (Kendall et al 2015) CNN architecture, with minor changes, for regressing the position of a ground robot using a compound image, consisting of images from four non-stereo monocular cameras mounted on the robotic platform. The training of the network is done end-to-end without the need for any special feature engineering to handle the compound image input. Our results show that there is significant advantage to using a compound image obtained from multiple cameras with non-overlapping field of view (non-stereo) as compared to using images from single cameras for image based localization. The compound-image based training yields median accuracy of 12 cm in an indoor environment which is at least twice as good as the results obtained using the same network trained on monocular image inputs.

In this paper we study and implement real-time adaptation schemes for video encoding and channel selection that work in tandem to facilitate HD video streaming for secondary users in a dynamic spectrum access network. Out-of-band feedbacks on instantaneous pathloss of the signal between the transmitter and the receiver, the received signal strength indicator (RSSI) at the receiver and the quality of the reconstructed video are used to continuously determine the most apt encoding parameters. At the same time, the radio transmitter continuously adjusts the channel parameters (i.e., center frequency and channel bandwidth) based on the transmission activities of the primary users who have prioritized rights on these channels. We consider the physical limitations of the encoder along with the channel statistics to determine when to change the encoder parameters and when to switch to a new channel. We propose a multi-level threshold based mechanism to find the optimal number of encoding bit rates. We also propose a threshold based algorithm to find the best available channel between the transmitter-receiver pair. We validate our theoretical propositions on an indoor testbed using software defined radios (SDRs) and the GNU Radio suite. Live video was captured, encoded using open source H.264 software libraries, streamed using GStreamer and transmitted over the 915 MHz ISM bands with omnidirectional antennas. For the SDRs, we chose the universal software radio peripheral (USRP) B210s from Ettus Research and use them as the transmitter and the receiver. A third B210 was used to sense the energy levels on all the channels to detect the presence of primary transmissions. GNU Radio was used to build the signal processing pipeline, both for the transmitter and the receiver. We use PSNR and SSIM to measure the video quality and report experimental results that show that: (i) the video encoder and the USRP transmitter-receiver pair are able to adapt to the changing RF conditions, (ii) the adaptation schemes yield better video quality than non-adaptive schemes, and (iii) the USRPs can switch the channels fast enough allowing uninterrupted HD video streaming even when primary users preempt the secondary user's transmission.

The influence class of network problems models the propagation of influence (an abstraction of cascading beliefs, behaviors, or physical phenomena) in a network. Such problems have applications in social networks, electrical networks, computer networks, viral spreading, and so on. These types of networks have also been studied through the lens of critical arcs detection; that is, which arcs (edges) are the most important for maintaining some property of the network (e.g., connectivity). We introduce a new class of problems at the intersection of these two models. Specifically, given a set of seed nodes and the linear threshold influence propagation model, our work proposes to determine which arcs (e.g., relationships in a social network or communication pathways in a telecommunication network) are most critical to the influence propagation process. We prove NP‐hardness of the problem. Time‐dependent and time‐independent mixed‐integer programming (MIP) models are introduced. Insights gleaned from MIP solutions leads to the development of an improved MIP‐based exact algorithm rooted in the idea of diffusion expansion. A heuristic based upon a new centrality measure is also proposed, and computational results are presented. © 2017 Wiley Periodicals, Inc. NETWORKS, Vol. 71(4), 412–431 2018

We propose a two-stage stochastic programming framework for designing or identifying “resilient,” or “reparable” structures in graphs whose topology may undergo a stochastic transformation. The reparability of a subgraph satisfying a given property is defined in terms of a budget constraint, which allows for a prescribed number of vertices to be added to or removed from the subgraph so as to restore its structural properties after the observation of random changes to the graph's set of edges. A two-stage stochastic programming model is formulated and is shown to be -complete for a broad range of graph-theoretical properties that the resilient subgraph is required to satisfy. A general combinatorial branch-and-bound algorithm is developed, and its computational performance is illustrated on the example of a two-stage stochastic maximum clique problem. © 2017 Wiley Periodicals, Inc. NETWORKS, Vol. 69(2), 189–204 2017

Truth Finding is the problem of determining correct information from several conflicting sources and is required for data aggregation. Existing algorithms solve the problem by simultaneously estimating source qualities and fact confidences, working on either numeric or non-numeric data. However, in practice, datasets are a mixture of several different data types. In this work we present a unified framework for finding truth from a collection of conflicting, authoritative sources. We assume that a small subset of independent reliable sources are selected by a preprocessing step. We formulate truth finding as an outlier removal problem, by modeling the similarities between the values reported by these sources. Our algorithm works in two stages: it first models the similarity graph between the sources and then finds the truth by invoking an outlier removal algorithm. We report experiments on several datasets including results for fixing records in Open Library; an open, editable library catalog.

In this paper, we present a coarse and passive localization system for GPS-denied environments. Our system is based on a cheap software defined radio (SDR) costing less than $25, which is used to listen to broadcast signals from local FM Radio stations. We show that the hardware and associated algorithms are capable of localizations with average errors of less than 5 miles, without requiring a fingerprinting or crowd sourcing approach. The algorithm is based on a large-scale simulation based map building phase and a query phase. We use 924 power spectra distributed over more than 200 miles for experimentally validating our query algorithm. Our average localization error is less than 5 miles.

This paper considers the minimum k-connected d-dominating set problem, which is a fault-tolerant generalization of the minimum connected dominating set (MCDS) problem. Three integer programming formulations based on vertex cuts are proposed (depending on whether d < k, d = k, or d > k) and their integer hulls are studied. The separation problem for the vertex-cut inequalities is a weighted vertex-connectivity problem and is polytime solvable, meaning that the LP relaxation can be solved in polytime despite having exponentially many constraints. A new class of valid inequalities—r-robust vertex-cut inequalities—is introduced and is shown to induce exponentially many facets. Finally, a lazy-constraint approach is shown to compare favorably with existing approaches for the MCDS problem (the case k = d = 1), and is in fact the fastest in literature for standard test instances. A key subroutine is an algorithm for finding an inclusion-wise minimal vertex cut in linear time. Computational results for (k, d) = (2,1), (2,2), (3,3), (4,4) are provided as well.

We present an efficient scenario decomposition algorithm for solving large-scale convex stochastic programming problems that involve a particular class of downside risk measures. The considered risk functionals encompass coherent and convex measures of risk that can be represented as an infimal convolution of a convex certainty equivalent, and include well-known measures, such as conditional value-at-risk, as special cases. The resulting structure of the feasible set is then exploited via iterative solving of relaxed problems, and it is shown that the number of iterations is bounded by a parameter that depends on the problem size. The computational performance of the developed scenario decomposition method is illustrated on portfolio optimization problems involving two families of nonlinear measures of risk, the higher-moment coherent risk measures, and log-exponential convex risk measures. It is demonstrated that for large-scale nonlinear problems the proposed approach can provide up to an order-of-magnitude improvement in computational time in comparison to state-of-the-art solvers, such as CPLEX, Gurobi, and MOSEK.

This study considers a class of critical node detection problems that involves minimization of a distance‐based connectivity measure of a given unweighted graph via the removal of a subset of nodes (referred to as critical nodes) subject to a budgetary constraint. The distance‐based connectivity measure of a graph is assumed to be a function of the actual pairwise distances between nodes in the remaining graph (e.g., graph efficiency, Harary index, characteristic path length, residual closeness) rather than simply whether nodes are connected or not, a typical assumption in the literature. We derive linear integer programming (IP) formulations, along with additional enhancements, aimed at improving the performance of standard solvers. For handling larger instances, we develop an effective exact algorithm that iteratively solves a series of simpler IPs to obtain an optimal solution for the original problem. The edge‐weighted generalization is also considered, which results in some interesting implications for distance‐based clique relaxations, namely, ‐clubs. Finally, we conduct extensive computational experiments with real‐world and randomly generated network instances under various settings that reveal interesting insights and demonstrate the advantages and limitations of the proposed approach. In particular, one important conclusion of our work is that vulnerability of real‐world networks to targeted attacks can be significantly more pronounced than what can be estimated by centrality‐based heuristic methods commonly used in the literature. © 2015 Wiley Periodicals, Inc. NETWORKS, Vol. 66(3), 170–195 2015

We present a novel framework, namely, accelerated alternating direction method of multipliers (AADMM), for acceleration of linearized ADMM. The basic idea of AADMM is to incorporate a multistep acceleration scheme into linearized ADMM. We demonstrate that for solving a class of convex composite optimization with linear constraints, the rate of convergence of AADMM is better than that of linearized ADMM, in terms of their dependence on the Lipschitz constant of the smooth component. Moreover, AADMM is capable of dealing with the situation when the feasible region is unbounded, as long as the corresponding saddle point problem has a solution. A backtracking algorithm is also proposed for practical performance.

We consider the problem of assigning a group of users with different rate requirements to a set of frequency bands, which may have different bandwidths, when the users access the channel via carrier-sense multiple access (CSMA). For example, in systems that employ dynamic spectrum access (DSA), secondary users may interleave their transmissions into space-time-frequency slots left open by primary users. The set of primary users that are active at a given time may leave a set of available channels that have unequal bandwidths. The use of CSMA by the secondary users reduces the need to coordinate transmissions among the users allocated to a particular frequency band, but it also results in potential collisions, which reduce the overall data rate that can be accommodated in the band. This is especially true when the presence of a finite sensing delay is considered. In this paper, we consider the layer problem of allocating users to the available frequency bands to minimize the bandwidth used (to accommodate other groups of secondary users), while taking into account the CSMA interactions of assigning multiple users to a band. We formulate this as a new form of bin packing problem, in which the size of the bin depends on the number of users that are assigned to the bin. A near optimal solution to this problem is found numerically using the Gurobi solver, and the performance is compared with the suboptimal first-fit algorithm, which has complexity that is appropriate for online implementation. Simulation results are provided to compare the optimal and online algorithms in terms of efficiency in allocating the bandwidth to the users and their complexity.

We study the minimum vertex blocker clique problem (VBCP),1 which is to remove a subset of vertices of minimum cardinality in a weighted undirected graph, such that the maximum weight of a clique in the remaining graph is bounded above by a given integer r ≥ 1. Cliques are among the most popular concepts used to model cohesive clusters in different graph‐based applications, such as social, biological, and communication networks. The general case of VBCP on weighted graphs is known to be NP‐hard, and we show that the special case on unweighted graphs is also NP‐hard for any fixed integer r ≥ 1. We present an analytical lower bound on the cardinality of an optimal solution to VBCP, as well as formulate VBCP as a linear 0–1 program with an exponential number of constraints. Facet‐inducing inequalities for the convex hull of feasible solutions to VBCP are also identified. Furthermore, we develop the first exact algorithm for solving VBCP, which solves the proposed formulation by using a row generation approach. Computational results obtained by utilizing this algorithm on a test‐bed of randomly generated instances are also provided. © 2014 Wiley Periodicals, Inc. NETWORKS, Vol. 64(1), 48–64 2014

In this report we describe the progress in developing a control architecture for human-in-the-loop wide area search munitions to reduce operator errors in the presence of unreliable automation and operator cognitive limitations. An optimal input tracking controller with adaptive automation uncertainty compensation and real-time workload assessment is developed to improve the system performance. Extensive simulations based on the experimental data involving 12 subjects demonstrate effectiveness of the presented controller.

Autonomous mobile robots, such as unmanned aerial or ground vehicles, will play important roles in future military and commercial applications. Future systems will require the robots to communicate with each other on a peer-to-peer basis over a wireless link and are able to form a networked system such as an ad hoc network. In this paper, we consider a problem of optimizing the topology of the network for such systems. Our solution consists of two steps. We first develop a novel method to select a network tree topology from arbitrary initial connected network topology. Second, we develop a new optimization algorithm to reconfigure the network, starting from the tree found in the first step, while maintaining the connectivity to minimize the aggregate traffic under the constraints on the total number of hops that the mobile robots may move. We present simulation results to compare the performance of our algorithms.