Transmission line parameters play a key role when accurately modeling the power network. However, several factors such as environmental conditions, errors in manufacturing data, and uninformed network changes can introduce errors to the line parameter information at the control center. In this paper, we leverage the sparse nature of parameter errors and propose an iterative greedy algorithm for nonlinear sparse error correction of parameter errors based on SCADA measurements. A benchmark approach of sparse error correction requires long computation time due to the slow convergence of the iterative reweighted least squares algorithm, with the issue getting aggravated as the power network scales. In the experiments using the IEEE 57-bus and IEEE 118-bus test cases, we demonstrate that our approach outperforms the benchmarks in terms of error correction performance and computation latency.

This paper considers the problem of performing post-attack forensic analysis for a test-time attack on a machine learning model. A test-time attack can be represented as a mapping that receives a benign test example as the input and outputs a falsified version of it. Given a set of attacked examples in the post-attack time, our objective is to identify the correct attack mapping among the collection of candidate attack strategies with diverse objectives and constraints. We present an attack mapping identification method that utilizes a pre-attack example recovery mechanism as a feature extraction method. Using numerical experiments, we demonstrate the effectiveness of the proposed approach in detecting the correct attack mapping among a number of different candidate attack strategies.

We consider a Bayesian binary detection problem in the presence of an adversary. The adversary is assumed to be capable of falsifying the observation, under a norm constraint, such that a detector will make a decision based on the falsified observation. We consider a sequential game between the adversary and the defender. The defender first determines the detection rule. Then, given the knowledge of the detection rule, the adversary determines the attack strategy to falsify the observation such that the probability of detection based on the falsified observation can be minimized. We prove that there exists a deterministic, optimal detection rule that the defender can use to achieve the maximum detection probability. We further prove that the maximum detection probability can be achieved by employing a deterministic detection rule where the decision is made based on proximity of the observation to a certain set.

We consider the problem of designing a robust classifier in the presence of an adversary who aims to degrade classification performance by elaborately falsifying the test instance. We propose a model-agnostic defense approach wherein the true class label of the falsified instance is inferred by analyzing its proximity to each class as measured based on class-conditional data distributions. We present a k-nearest neighbors type approach to perform a sample-based approximation of the aforementioned probabilistic proximity analysis. The proposed approach is evaluated on three different real-world datasets in a game-theoretic setting, in which the adversary is assumed to optimize the attack design against the employed defense approach. In the game-theoretic evaluation, the proposed defense approach significantly outperforms benchmarks in various attack scenarios, demonstrating its efficacy against optimally designed attacks.

The hyperspectral unmixing problem arises in remote sensing, chemometrics, and biomedical engineering applications. The spectral signature of a single pixel in a hyperspectral cube can be represented as a non-negative combination of non-negative signatures from various materials contained in the physical region corresponding to the pixel (linear mixing). A less studied problem is associated with foreground extraction in an intimate (nonlinear) mixing model. We introduce a framework for foreground signature extraction based on a proposed patch model. We introduce identifiability conditions for the single and multiple patch cases. Using these conditions, we present an algorithm for the identifiable recovery of foreground signatures. Numerical experiments on real and synthetic data illustrate the efficacy of the proposed approach.

The problem of making the classifier design resilient to test data falsification is considered. In the literature, a few countermeasures have been proposed to defend machine learning algorithms against test data falsification, but a common assumption employed therein is that feature entries of test data are equally vulnerable to falsification. When test data entries consist of data collected from various sources such as different types of sensor devices, vulnerability levels of data entries to falsification attacks can differ significantly depending on how data creation and transmission procedures are secured. In this paper, we present an attack-cost-aware adversarial learning framework that takes into account the (potentially inhomogeneous) vulnerability characteristics of test data entries in designing an attack-resilient classifier. We demonstrate the efficacy of the proposed approach using experiments with the MNIST handwritten digit database.

The problem of correcting power network parameters and topology using multi-period SCADA measurements is considered. Starting from the current knowledge of parameter values, we formulate the parameter correction problem as a sparse unsupervised regression problem by exploiting the sparsity of the parameter errors. The advantage of the proposed approach is that it can localize and estimate parameter errors at the same time; there is no need for prior knowledge of error locations. Furthermore, the approach can be adapted to correct sparse errors in both parameters and topology simultaneously. We present an iterative parameter correction algorithm and demonstrate its efficacy using the IEEE 14-bus test case.

We study sparse gross error correction for state estimation in a non-linear sensing system. We consider a practical assumption that gross errors are sparse, and their locations tend to be invariant over a few consecutive measurement periods. Under the assumption, a robust state estimation and error correction algorithm using multiple measurement vectors is proposed based on local linear approximation of the nonlinear measurement model. Unlike existing approaches in the literature, the proposed method ensures that the estimated gross error locations are such that system is observable, i.e., the system state is uniquely identifiable. The proposed method was applied for power system AC state estimation of the IEEE 14-bus network and outperformed benchmark techniques.

Recognizing and localizing a recurring pattern is a problem with a variety of applications such as classification and localization of home appliances from their activation signals and estimating the relative alignment between records of a natural repetitive electrocardiography (ECG) signals in Bio-medical data. Most common approaches for recognizing a recurring pattern are generative and focus on discovering and capturing the characteristics of the recurring pattern. One limitation of such generative approaches is that they are more sensitive to variations of the recurring signal. In this paper, we present a discriminative approach for finding a recurring pattern and localizing it within a collection of signals. We evaluate and compare our method to a generative approach [1] on both synthetic data and real-world home appliance data.

A new mechanism aimed at misleading a power system control center about the source of a data attack is proposed. As a man-in-the-middle state attack, a data framing attack is proposed to exploit the bad data detection and identification mechanisms currently in use at most control centers. In particular, the proposed attack frames meters that are providing correct data as sources of bad data such that the control center will remove useful measurements that would otherwise be used by the state estimator. The optimal design of a data framing attack is formulated as a quadratically constrained quadratic program. It is shown that the proposed attack is capable of perturbing the power system state estimate by an arbitrary degree controlling only half of a critical set of measurements that are needed to make a system unobservable. Implications of this attack on power system operations are discussed, and the attack performance is evaluated using benchmark systems.

Covert data attacks on the network topology of a smart grid is considered. In a so-called man-in-the-middle attack, an adversary alters data from certain meters and network switches to mislead the control center with an incorrect network topology while avoiding detections by the control center. A necessary and sufficient condition for the existence of an undetectable attack is obtained for strong adversaries who can observe all meter and network data. For weak adversaries with only local information, a heuristic method of undetectable attack is proposed. Countermeasures to prevent undetectable attacks are also considered. It is shown that undetectable attacks do not exist if a set of meters satisfying a certain branch covering property are protected. The proposed attacks are tested with IEEE 14-bus and IEEE 118-bus system, and their effect on real-time locational marginal pricing is examined.

The problem of localizing in-band wormhole tunnels inMANETs is considered. In an in-band wormhole attack, colluding attackers use a covert tunnel to create the illusion that two remote network regions are directly connected. This apparent shortcut in the topology attracts traffic which the attackers can then control.