Adversarial attacks have consistently posed a critical challenge to the reliability of deep learning models. This warrants heightened concern towards malicious cyber attacks in sensitive application domains of cybersecurity where such learning models are relied upon to detect and mitigate adversarial threats. In this work, we present a method to improve the robustness of Network Intrusion Detection System (NIDS) models by leveraging a game theoretical adversarial training algorithm with a variational adversary to generate optimal perturbations causing misclassifications. Subsequently, we show that integrating such adversarially manipulated data samples into the training algorithm leads to an improvement in the resilience of the NIDS classifiers against such attacks. To validate our method, we also simulate cyber attacks in a Software Defined Networking (SDN) environment, generating synthetic traffic data for adversarial training. We detail the setup for this simulation, providing a reproducible framework for adversarial training in security contexts. This complements our experiments on widely used benchmark datasets for network intrusion detection systems such as CSE-CIC-IDS2018 and NSL-KDD. We also assess the generality of the proposed method by applying it to the emerging Kolmogorov Arnold Networks (KANs). Our results confirm that the game theory based adversarial training algorithm significantly enhances the robustness of KANs, highlighting the value of our approach.

Identifying causal drivers in multivariate time-series data is central to finance, climate science, and other domains where interactions are nonlinear, high-dimensional, and noisy. Standard Granger-causality tests, based on linear vector-autoregressive models and F-statistics, fail to capture such complexity, and do not scale to modern data volumes. We introduce a unified deep-learning framework that 1) models nonlinear lag dependencies with neural networks trained under empirical-risk minimization, 2) quantifies multivariate causal strength via a generalized-variance criterion, 3) enhances interpretability through Bayesian-network feature selection, and wavelet-based multiscale analysis, and 4) implements mean squared error and quantile-based loss function to uncover causal effects. Extensive experiments on S&P 500 and NIFTY 50 datasets demonstrate that the implemented techniques effectively uncover Granger-causal relationships, yielding significant improvements, up to 82%, in predictive performance for the multivariate feature-set. Successful implementation of a diverse range of techniques further confirms that the proposed causal detection framework provides a scalable, domain-agnostic approach to causal feature extraction.

In this chapter, we explore adversarial attack surfaces. We examine how they can exploit vulnerabilities in machine learning and how to make learning algorithms robust to attacks on security and privacy of the learning system. To explore the vulnerabilities, we can simulate various model training processes under a range of various attack scenarios in supervised and unsupervised settings. Each attack strategy is assumed to be formulated by an intelligent adversary that is capable of either feature manipulation, label manipulation, or both. The optimal attack policy of the adversaries is determined by the solution for optimization problems that output the adversarial data. We can then apply the knowledge that we learned to improve and reinforce the learning procedure so as to better defend against attacks. The sensitivity analysis summarized in this chapter can be used to develop computational algorithms for optimization objectives and statistical inferences in adversarial learning algorithm's capacity for randomization, discrimination, reliability, and learnability. It creates research pathways into robustness, fairness, explainability, and transparency of machine learning models.

While adversarial examples (AEs) or adversarial perturbations (APs) are usually treated as a security risk up to date, they can also serve as privacy protection tools when facing deep learning-based privacy attacks. This chapter will first introduce a privacy model for visual data, one of the most important types of data in deep learning applications. Then we will discuss AP-based privacy protection mechanisms that incorporate different levels of privacy. While the research on this topic is still in its infancy stage, this chapter will overview the state-of-the-art works and shed light on future research.

During the past decades, deep neural networks (DNNs) have shown great success in a wide range of applications, including image classification in the computer vision (CV) domain [263, 282, 558] and text recognition in the natural language processing (NLP) field [157, 276]. However, recent researches have shown that DNNs are immensely brittle toward adversarial examples primarily in the image domain [228, 589]. For example, Goodfellow et al. [228] demonstrated that adding nearly zero noises to a panda image can mislead the GoogLeNet to incorrect label (gibbon) with high confidence (99.3%). This phenomenon raises great concern about DNNs security implementation and attracts much attention in the CV community since 2014. In the literature, numerous approaches have been proposed to generate adversarial examples to attack DNNs (aka, the attack branch) and design corresponding mechanisms to defense these potential attacks (aka, the defense branch). In this chapter, we focus on the adversarial attack direction to craft high-quality adversarial examples in both the CV domain and the NLP domain.

A critical challenge in machine learning is the vulnerability of learning models in defending attacks from malicious adversaries. In this research, we propose game theoretical learning between a variational adversary and a Convolutional Neural Network (CNN), participating in a variable-sum two-player sequential Stackelberg game. Our adversary manipulates the input data distribution to make the CNN misclassify the manipulated data. Our ideal adversarial manipulation is a minimum change to the data which yet is large enough to mislead the CNNs. We propose an optimization procedure to find optimal adversarial manipulations by solving for the Nash equilibrium of the Stackelberg game. Specifically, the adversary's payoff function depends on the data manipulation which is determined by a Variational Autoencoder, while the CNN classifier's payoff functions are evaluated by misclassification errors. The optimization of our adversarial manipulations is defined by Alternating Least Squares and Simulated Annealing. Experimental results demonstrate that our game-theoretic manipulations are able to mislead CNNs that are well trained on the original data as well as on data generated by other models. We then let the CNNs to incorporate our manipulated data which leads to secure classifiers that are empirically the most robust in defending various types of adversarial attacks.

With the number of users of social media and web platforms increasing day-by-day in recent years, cyberbullying has become a ubiquitous problem on the internet. Controlling and moderating these social media platforms manually for online abuse and cyberbullying has become a very challenging task. This paper proposes a Recurrent Neural Network (RNN) based approach for the identification and classification of cyberbullying posts. In highly imbalanced input data, a Tomek Links approach does under-sampling to reduce the data imbalance and remove ambiguities in class labelling. Further, the proposed classification model uses Max-Pooling in combination with Bi-directional Long Short-Term Memory (LSTM) network and attention layers. The proposed model is evaluated using Wikipedia datasets to establish the effectiveness of identifying and classifying cyberbullying posts. The extensive experimental results show that our approach performs well in comparison to competing approaches in terms of precision, recall, with F1 score as 0.89, 0.86 and 0.88, respectively.

We develop an adversarial learning algorithm for supervised classification in general and Convolutional Neural Networks (CNN) in particular. The algorithm's objective is to produce small changes to the data distribution defined over positive and negative class labels so that the resulting data distribution is misclassified by the CNN. The theoretical goal is to determine a manipulating change on the input data that finds learner decision boundaries where many positive labels become negative labels. Then we propose a CNN which is secure against such unforeseen changes in data. The algorithm generates adversarial manipulations by formulating a multiplayer stochastic game targeting the classification performance of the CNN. The multiplayer stochastic game is expressed in terms of multiple two-player sequential games. Each game consists of interactions between two players-an intelligent adversary and the learner CNN-such that a player's payoff function increases with interactions. Following the convergence of a sequential noncooperative Stackelberg game, each two-player game is solved for the Nash equilibrium. The Nash equilibrium finds a pair of strategies (learner weights and evolutionary operations) from which there is no incentive for either learner or adversary to deviate. We then retrain the learner over all the adversarial manipulations generated by multiple players to propose a secure CNN which is robust to subsequent adversarial data manipulations. The adversarial data and corresponding CNN performance is evaluated on MNIST handwritten digits data. The results suggest that game theory and evolutionary algorithms are very effective in securing deep learning models against performance vulnerabilities simulated as attack scenarios from multiple adversaries.

Deep networks are vulnerable to adversarial attacks from malicious adversaries. Currently, many adversarial learning algorithms are designed to exploit such vulnerabilities in deep networks. These methods focus on attacking and retraining deep networks with adversarial examples to do either feature manipulation or label manipulation or both. In this paper, we propose a new adversarial learning algorithm for finding adversarial manipulations to deep networks. We formulate adversaries who optimize game-theoretic payoff functions on deep networks doing multi-label classifications. We model the interactions between a classifier and an adversary from a game-theoretic perspective and formulate their strategies into a Stackelberg game associated with a two-player problem. Then we design algorithms to solve for the Nash equilibrium, which is a pair of strategies from which there is no incentive for either the classifier or the adversary to deviate. In designing attack scenarios, the adversary's objective is to deliberately make small changes to test data such that attacked samples are undetected. Our results illustrate that game-theoretic modelling is significantly effective in securing deep learning models against performance vulnerabilities attached by intelligent adversaries.