Search

The proliferation of novel attacks and growing amounts of data has caused practitioners in the field of network intrusion detection to constantly work towards keeping up with this evolving adversarial landscape. Researchers have been seeking to harness deep learning techniques in efforts to detect zero-day attacks and allow network intrusion detection systems to more efficiently alert network operators. The technique outlined in this work uses a one-class training process to shape autoencoder feature residuals for the effective detection of network attacks. Compared to an original set of input features, we show that autoencoder feature residuals are a suitable replacement, and often perform at least as well as the original feature set. This quality allows autoencoder feature residuals to prevent the need for extensive feature engineering without reducing classification performance. Additionally, it is found that without generating new data compared to an original feature set, using autoencoder feature residuals often improves classifier performance. Practical side effects from using autoencoder feature residuals emerge by analyzing the potential data compression benefits they provide. [ABSTRACT FROM AUTHOR]

For many important network types, physical coordinate systems and physical distances are either difficult to discern or inapplicable. Accordingly, coordinate systems and characterizations based on hop-distance measurements, such as Topology Preserving Maps (TPMs) and Virtual-Coordinate (VC) systems are attractive alternatives to geographic coordinates for many network algorithms. We present an approach to recover geometric and topological properties of a network with a small set of distance measurements. The approach is a combination of shortest path (often called geodesic) recovery concepts and low-rank matrix completion, generalized to the case of hop-distances in graphs. Results for sensor networks embedded in 2-D and 3-D spaces as well as for social networks indicate that the method can accurately capture the network connectivity with a small set of measurements. TPM generation can now be based on various context appropriate measurements or VC systems instead of distances to a set of global anchors. The proposed method is a significant generalization that allows the topology to be extracted from a random set of graph shortest paths, making it applicable in contexts such as social networks where VC generation may not be possible. [ABSTRACT FROM AUTHOR]

In computer networks, denial-of-service (DoS) attacks attempt to make computers or network resources unavailable for their intended use. DoS attacks are difficult to detect and mitigate since they normally do not attempt to access the private data of their intended victim, but rather intend to disrupt the publicly available resources their victims provide. This article discusses methods for detecting and mitigating DoS attacks with a focus on techniques that leverage machine learning algorithms. Such algorithms promise to: (a) detect when computer services are being used in an adversarial fashion, (b) separate network traffic into nominal and anomalous components, and (c) provide opportunities for mitigating the attacks while maintaining the integrity of the effected services. The key ingredient of the ideas presented here is the use of correlations and dependencies in computer access patterns, and the larger context in which they exist, to separate the “wheat” – the real users of the services – from the “chaff” –the perpetrators who are attempting to disrupt the services. [ABSTRACT FROM AUTHOR]

The National Aeronautics and Space Administration (NASA) is in the midst of defining and developing the future space and ground architecture for the coming decades to return science and exploration discovery data back to investigators on Earth. Optimizing the data return from these missions requires planning, design, standards, and operations coordinated from formulation and development throughout the mission. The use of automation enhanced by cognition and machine learning are potential methods for optimizing data return, reducing costs of operations, and helping manage the complexity of the automated systems. In this article, we discuss the potential role of machine learning in the linkto- link aspect of the communication systems. An experiment using NASA's Space Communication and Navigation Testbed onboard the International Space Station and the ground station located at NASA John H. Glenn Research Center demonstrates for the first time the benefits and challenges of applying machine learning to space links in the actual flight environment. The experiment used machine learning decisions to configure a space link from the ISS-based testbed to the ground station to achieve multiple objectives related to data throughput, bandwidth, and power. Aspects of the specific neural-network-based reinforcement learning algorithm formation and on-orbit testing are discussed. [ABSTRACT FROM AUTHOR]

Chemical sensors play an important role in a variety of civilian and military domains. In these contexts, the ability to accurately and quickly identify chemical agents is of utmost importance. In practice, constraints on physical footprint, power consumption, ease of use, and time required for accurate detection often restrict the utility of sensors, particularly in remote and isolated regions. One solution to address this problem is the engineering of advanced signal processing techniques, which decrease the time required for accurate detection. This allows software to facilitate the construction of hardware that meet stringent power and concept of operations guidelines. In this paper, we propose the Kalman filter as a preprocessing technique applicable to chemical sensor time series data for downstream machine learning. Using data collected from a sensor array of multiple unique polymer-graphene nanoplatelet coated electrodes, we show accurate and early detection of both organophosphates and interferents is improved when the Kalman filter is used as a preprocessing technique. In particular, within two seconds of analyte exposure to the sensor array, classification using Kalman filtered first derivative estimates achieve an error of less than 10%. By comparison, the non-Kalman filtered data set has a classification error rate above 40% within this time. An advantage of our approach is classification does not depend on a set parameter, such as maximum resistance change, or a pre-determined exposure time, and which allows rapid classification immediately after analyte introduction. [ABSTRACT FROM AUTHOR]

Future spacecraft communication subsystems will potentially benefit from software-defined radios controlled by artificial intelligence algorithms. In this paper, we propose a novel radio resource allocation algorithm leveraging multiobjective reinforcement learning and artificial neural network ensembles able to manage available resources and conflicting mission-based goals. The uncertainty in the performance of thousands of possible radio parameter combinations and the dynamic behavior of the radio channel over time producing a continuous multidimensional state–action space requires a fixed-size memory continuous state–action mapping instead of the traditional discrete mapping. In addition, actions need to be decoupled from states in order to allow for online learning, performance monitoring, and resource allocation prediction. The proposed approach leverages the authors’ previous research on constraining decisions predicted to have poor performance through ”virtual environment exploration.” The simulation results show the performance for different communication mission profiles, and accuracy benchmarks are provided for the future research reference. The proposed approach constitutes part of the core cognitive engine proof-of-concept delivered to the NASA John H. Glenn Research Center’s SCaN Testbed radios on-board the International Space Station. [ABSTRACT FROM AUTHOR]

We compare two mathematical theories that address duality between cycles and vertex-cuts of graphs in geometric settings. First, we propose a rigorous definition of a new type of graph, vector graphs. The special case of R²-vector graphs matches the intuitive notion of drawing graphs with edges taken as vectors. This leads to a discussion of Kirchhoff graphs, as originally presented by Fehribach, which can be defined independent of any matrix relations. In particular, we present simple cases in which vector graphs are guaranteed to be Kirchhoff or non-Kirchhoff. Next, we review Maxwell's method of drawing reciprocal figures as he presented in 1864, using modern mathematical language. We then demonstrate cases in which R²-vector graphs defined from Maxwell reciprocals are "dual" Kirchhoff graphs. Given an example in which Maxwell's theories are not sufficient to define vector graphs, we begin to explore other methods of developing dual Kirchhoff graphs. [ABSTRACT FROM AUTHOR]

A theory and algorithm for detecting and classifying weak, distributed patterns in network data is presented. The patterns we consider are anomalous temporal correlations between signals recorded at sensor nodes in a network. We use robust matrix completion and second order analysis to detect distributed patterns that are not discernible at the level of individual sensors. When viewed independently, the data at each node cannot provide a definitive determination of the underlying pattern, but when fused with data from across the network the relevant patterns emerge. We are specifically interested in detecting weak patterns in computer networks where the nodes (terminals, routers, servers, etc.) are sensors that provide measurements (of packet rates, user activity, central processing unit usage, etc.). The approach is applicable to many other types of sensor networks including wireless networks, mobile sensor networks, and social networks where correlated phenomena are of interest. [ABSTRACT FROM PUBLISHER]

We consider the problem of minimizing the energy of an inextensible elastic strut with length 1 subject to an imposed twist angle and force. In a standard calculus of variations approach, one first locates equilibria by solving the Euler-Lagrange ODE with boundary conditions at arclength values 0 and 1. Then one classifies each equilibrium by counting conjugate points, with local minima corresponding to equilibria with no conjugate points. These conjugate points are arclength values σ ≤ 1 at which a second ODE (the Jacobi equation) has a solution vanishing at 0 and σ. Finding conjugate points normally involves the numerical solution of a set of initial value problems for the Jacobi equation. For problems involving a parameter λ, such as the force or twist angle in the elastic strut, this computation must be repeated for every value of λ of interest. Here we present an alternative approach that takes advantage of the presence of a parameter λ. Rather than search for conjugate points σ ≤ 1 at a fixed value of λ, we search for a set of special parameter values λ[SUBm] (with corresponding Jacobi solution ζ[SUPm]) for which σ = 1 is a conjugate point. We show that, under appropriate assumptions, the index of an equilibrium at any λ equals the number of these ζ[SUPm] for which 〈ζ[SUPm],Sζ,[SUPm]〉, < 0 where S is the Jacobi differential operator at λ. This computation is particularly simple when λ appears linearly in S. We apply this approach to the elastic strut, in which the force appears linearly in S, and, as a result, we locate the conjugate points for any twisted unbuckled rod configuration without resorting to numerical solution of differential equations. In addition, we numerically compute two-dimensional sheets of buckled equilibria (as the two parameters of force and twist are varied) via a coordinated family of one-dimensional parameter... [ABSTRACT FROM AUTHOR]