Energy monitoring and appliance identification are critical for addressing energy crisis and environmental pollution. However, variations in appliance load profiles across different households and locations pose significant challenges to accurate identification. Moreover, existing models often assume access to sufficient labeled training data, which can be costly or impractical to obtain. To overcome these challenges and enhance cross-domain generalization, we propose a vision transformer-based unsupervised adversarial appliance identification model that incorporates domain adaptation with CORrelation ALignment (CORAL) loss. A pretrained Vision Transformer (ViT) is used as a feature extractor to jointly learn shared representations for both the label predictor and domain classifier, enabling the model to focus on domain-specific characteristics and mitigate domain shift. Low-resolution daily power consumption signals are segmented into 15-minute resolution time-series windows and transformed into Gramian Angular Difference Field (GADF) images to enhance feature extraction. The model is then adversarially trained on source and target domains using a negative-log likelihood loss augmented with correlation alignment to improve stability and feature alignment. We evaluated the model under three different scenarios: (i) source and target domains from different geographic regions, (ii) source and target domains from the same region but different datasets (utilities), and (iii) source and target domains from the same dataset but different houses. Experiments on three public datasets demonstrate substantial improvements, achieving macro F1 score ranging from 12.66% to 70.82% in scenario (i), 2.87% to 84.62% in scenario (ii), and 7.77% to 362.56% in scenario (iii). An ablation study further reveals that incorporating CORAL loss can achieve up to a 75.34% improvement in macro F1 score compared to domain adaptation without it. Additionally, the proposed model consistently outperforms direct time-series models that uses 1D raw signals. It also demonstrates superior scalability, robustness under various input conditions, and efficient computational complexity with faster training and inference times.

Due to the scalability of the stateless and compact JSON Web Tokens (JWT), they are increasingly used in securing modern applications to support single-sign-on context and establish trust in microservices or connected-devices architectures. However, currently JWT relies on traditional encryption algorithms that are based on integer factorization or discrete logarithm problems; this exposes JWT signatures to vulnerability when quantum computers become available. Recently, NIST has announced new standards which are quantum safe. In this paper, we investigate the usage of two quantum-resistant algorithms, which are NIST candidates of round 2 standardization process in 2019, to create and verify JWT signatures. Namely, we consider lattice-based schemes (DILITHIUM and qTESLA) and compare their performance against RSA for digital signature for different security levels defined by NIST. The results show that lattice-based digital signature schemes have better performance than RSA in terms of request per seconds and average response time. DILITHIUM has demonstrated the best performance while also having smaller performance loss when scaling the security level.

With the massive increase of online content, widespread of social media, the popularity of smartphones, and rise of security breaches, image forensics has attracted a lot of attention in the past two decades alongside the advancements in digital imaging and processing software. The goal is to be able to verify authenticity, ownership, and copyright of an image and detect changes to the original image. However, more sophisticated image manipulation software tools can use subtle anti‐forensics techniques (AFTs) to complicate and hinder detection. This leads security professionals and digital investigators to develop more robust forensics tools and counter solutions to defeat adversarial anti‐forensics and win the race. This survey study presents a comprehensive systematic overview of various anti‐forensics and anti‐AFTs that are proposed in the literature for digital image forensics. These techniques are thoroughly analysed based on various important characteristics and grouped into broad categories. This study also presents a bibliographic analysis of the‐state‐of‐the‐art publications in various venues. It assists junior researchers in multimedia security and related fields to understand the significance of existing techniques, research trends, and future directions.

With the increasing rate and catastrophic consequences of phishing attacks, research on anti-phishing solutions has gained growing importance in information security. Security risks may include information leakage, identity theft, financial loss and reputation sabotage. Raising human awareness is not a sufficient mitigation method and deploying complementary technical solutions is a crucial requirement. Although various approaches have been proposed in the literature, the design of efficient phishing detection models is a challenging task and the problem still lacks a complete solution. In this paper, we present a novel approach for detecting phishing websites based on probabilistic neural networks (PNNs). We also investigate the integration of PNN with K-medoids clustering to significantly reduce complexity without jeopardizing the detection accuracy. To assess the feasibility of the proposed approach, we conducted in-depth study to evaluate various performance measures on a publicly available data set composed of 11 055 phishing and benign websites. The experimental results show that more accurate models can be built and even with >40% reduction in the complexity, >97% accuracy can be achieved with low false errors.

Electronic mail has become the most popular, frequently-used and powerful medium for quicker personal and business communications. However, one of the common security issues and annoying problems faced by email users and organizations is receiving a large number of unsolicited email messages, known as spam emails, every day. A traditional countermeasure in most email systems nowadays is simple filtering mechanisms that can block or quarantine unwanted emails based on some keywords defined by the user. These filters require continual effort to keep them relevant and current with some extensions proposed to improve their performance. However, due to the gigantic volumes of received emails and the continual change in spamming techniques to bypass the implemented solutions, novel automated ideas and countermeasures need to be investigated. This paper explores a novel algorithm inspired by the immune system called dendritic cell algorithm (DCA). This algorithm is evaluated on a number of benchmark datasets to detect spam emails. The results demonstrate that this approach can be a promising solution for email classification and spam filtering.

In this work, a spread spectrum watermarking optimization algorithm is explored for digital color images using biobjective genetic algorithms and full-frame discrete-cosine transform. The aim of optimization is to generate the trade-off curve, a.k.a. optimal Pareto points, of watermark imperceptibility and robustness. The watermark imperceptibility is evaluated using the Structural SIMilarity (SSIM) index between the original image and the watermarked image whereas the watermark robustness is evaluated in terms of the Normalized Correlation Coefficient (NCC) between the original watermark and the recovered watermark. The watermarked image is susceptible to various types of attacks or processing distortions such as additive Gaussian noise, pepper-and-salt noise, JPEG compression, camera motion and median filtering. For the biobjective genetic algorithm, we used the fast elitist Non-dominated Sorting Genetic Algorithm (NSGA-II). We reviewed related work and investigated two color spaces (YCbCr and HSV) in addition to gray scale images where embedding is conducted in different frames and various distortions are applied before the extraction of the watermark. The results are compared for various cases under similar conditions.

Due to the increasing use of digital images in electronic systems, image processing is gaining considerable attention nowadays. In this paper, we investigate the ability of adaptive neuro-fuzzy inference system (ANFIS) in assessing the quality of digital images. This is implemented through comparison of the predicted and actual differential mean opinion score (DMOS). Several distinguishing features are extracted and adopted to construct computational classification models that predict the DMOS value. We found that for a 7-input ANFIS network, the predicted DMOS values for distorted images of blur type have a high linear correlation coefficient of 0.9937, a Spearman’s ranked correlation of 0.9902, and RMSE of 3.2%. Moreover, the predicted DMOS values for distorted images of JPEG 2000 compression type have a high linear correlation coefficient of 0.9944, a Spearman’s ranked correlation of 0.9902, and RMSE of 3.32%. This shows that combining the advantages of both neural network and fuzzy systems can be a promising approach for assessing the quality of digital images.

A boosting-based cascade for automatic decomposition of multiclass learning problems into several binary classification problems is presented. The proposed cascade structure uses a boosted classifier at each level and use a filtering process to reduce the problem size at each level. The method has been used for detecting malicious traffic patterns using a benchmark intrusion detection dataset. A comparison of the approach with four boosting-based multiclass learning algorithms is also provided on this dataset.

With the wide use of steganographic techniques, several security challenges emerge, e.g. criminals and network intruders can hide any information they want into legitimate multimedia data and exchange it over the Internet. This requires network designers and service providers to investigate new tools for detecting such misuse. In this paper, we explore a detection method based on neural network approach with Levenberg-Marquardt back propagation learning algorithm. This learning technique has been known to overcome the slow convergence of traditional back propagation and the instability problem of the steepest descent optimization method. We focus on digital images containing messages embedded by one of the recently proposed steganographic methods, known as pixel-value differencing. The idea is to analyze images before and after embedding to extract discriminating features and then build a neural network recognition model. The proposed approach is empirically evaluated and compared with four other machine-learning methods. The results show that more than 99% detection rate can be attained with very few false alarms.

This paper explores a novel application for reinforcement learning (RL) techniques to sequential mastery testing. In such systems, the goal is to classify each examined person, using the minimal number of test items, as master or non-master. Using RL, an intelligent agent autonomously learns from interactions to administer more informative and effective variable-length tests. Empirical results are also provided to evaluate the performance of the proposed approach as compared to two common approaches for variable-length testing (Bayesian decision and sequential probability ratio test) as well as to the fixed-length testing.

Unlike data encryption, steganography provides a crucial means for hiding confidential data into cover media so that the unauthorized person will not be aware of the existence of this data. This is particularly useful for protecting sensitive data that needs to be transmitted over a public access network such as the Internet. Several schemes have been proposed based on pixel value differencing with an attempt to increase the embedding capacity of secret data into digital images without significant loss in the perceived quality of the cover images. This paper aims to survey a number of methods that depend on or extend the pixel value differencing scheme. It also provides empirical comparison and discussion of the differences between the various methods.

Recently, several image cryptographic algorithms based on chaotic maps have been proposed. These algorithms differ in various aspects such as the type of chaotic maps used, methods used for confusion and diffusion, key size, and manipulating images as 2D or 3D. In this paper, a new image encryption algorithm based on chaotic maps is proposed. The algorithm is inspired by two bio-operations: crossover and mutation. Experimental results show that the proposed algorithm is capable of generating encrypted images with uniform distribution of the gray scale values and very low correlation coefficients of adjacent pixels, and is very sensitive to any change in the secret key values. We also compared the proposed algorithm run time and correlation coefficients for encrypted images with two other approaches.

We propose a multiclass hierarchical abductive learning classifier and apply it to improve the recognition rate of handwritten numerals while reduce the dimensionality of the feature space. For handwritten recognition, there are ten classes. Using 9 binary GMDH-based neural network models structured in a hierarchy has led to improving balance factor of the dataset for each classifier and improving the classification of handwritten numerals. It also has the advantage of removing the need to resolve classification ties that exist in other forms of combining a number of classifiers to solve a multiclass classification problem whether using one-versus-all or one-versus-one approaches. The proposed approach is empirically evaluated and compared with five other state-of-the-art machine learning classifiers using a publicly available dataset based on non-Gaussian topological features.

The cost estimates for receiving unsolicited commercial email messages, also known as spam, are threatening. Spam has serious negative impact on the usability of electronic mail and network resources. In addition, it provides a medium for distributing harmful code and/or offensive content. The work in this paper is motivated by the dramatic increase in the volume of spam traffic in recent years and the promising ability of ant colony optimization in data mining. Our goal is to develop an ant-colony based spam filter and to empirically evaluate its effectiveness in predicting spam messages. We also compare its performance to three other popular machine learning techniques: multi-layer perceptron, naive Bayes and Ripper classifiers. The preliminary results show that the developed model can be a remarkable alternative tool in filtering spam; yielding better accuracy with considerably smaller rule sets which highlight the important features in identifying the email category.

E-mail spam has become an epidemic problem that can negatively affect the usability of electronic mail as a communication means. Besides wasting users' time and effort to scan and delete the massive amount of junk e-mails received; it consumes network bandwidth and storage space, slows down e-mail servers, and provides a medium to distribute harmful and/or offensive content. Several machine learning approaches have been applied to this problem. In this paper, we explore a new approach based on fuzzy similarity that can automatically classify e-mail messages as spam or legitimate. We study its performance for various conjunction and disjunction operators for several datasets. The results are promising as compared with a naive Bayesian classifier. Classification accuracy above 97% and low false positive rates are achieved in many test cases.

Spam messages pose a major threat to the usability of electronic mail. Spam wastes time and money for network users and administrators, consumes network bandwidth and storage space, and slows down email servers. In addition, it provides a medium to distribute harmful code and/or offensive content. In this paper, we investigate the application of abductive learning in filtering out spam messages. We study the performance for various network models on the spambase dataset. Results reveal that classification accuracies of 91.7% can be achieved using only 10 out of the available 57 content attributes. The attributes are selected automatically by the abductive learning algorithm as the most effective feature subset, thus achieving approximately 6:1 data reduction. Comparison with other techniques such as multi-layer perceptrons and naive Bayesian classifiers show that the abductive learning approach can provide better spam detection accuracies, e.g. false positive rates as low as 5.9% while requiring much shorter training times.

In today's online connected world, almost all corporate networks use some form of perimeter firewalls to manage Internet connections and enforce a security policy at the corporate gateway. Although it can considerably enhance network security and protect business-critical information, a firewall with thousands of rules can become a bottleneck for network performance. The primary goal of this paper is to present a new rule order optimizer based on simulated annealing to find optimal configurations that minimize the average number of rule comparisons while preserving precedence relationships among disjoint rules. The proposed approach is evaluated and its effectiveness is compared with another approximate solution under several firewall configurations and policy profiles.

This paper presents a flexible evolutionary method for minimum-cost multipath constrained routing with load balancing problem in MPLS networks. The proposed solution approach combines genetic algorithms with linear multi-commodity flow to enhance the efficiency of the solution attained. The goal is to determine the distribution of traffic demands over a given capacitated network topology to minimize the routing cost while balancing loads on various links. The constraints that should be satisfied are the maximum hop count, the total number of virtual paths and the link capacities. This problem is a highly constrained multiobjective optimization for which exact optimization methods become helpless to deal with such complexity. Using a case study from the literature, the proposed approach is evaluated and compared with the standard genetic algorithm. We also show how the proposed approach can be used to determine approximate Pareto points and compare them with the exact Pareto front.

This paper addresses the application of genetic algorithms (GA) to the optimal topology design of MPLS networks. This problem is a highly constrained optimization problem for which exact solution approaches do not scale well. We first use a layered model and decompose the MPLS topology design as a set of linear programs. Then, we propose a heuristic approach based on genetic algorithms for solving them. Simulation results show that the proposed approach is effective and give optimal or close to optimal solutions for the tested cases.

An efficient channel allocation policy that prioritizes handoffs is an indispensable ingredient in future cellular networks in order to support multimedia traffic while ensuring quality of service requirements (QoS). In this paper we study the application of a reinforcement-learning algorithm to develop an alternative channel allocation scheme in mobile cellular networks that supports multiple heterogeneous traffic classes. The proposed scheme prioritizes handoff call requests over new calls and provides differentiated services for different traffic classes with diverse characteristics and quality of service requirements. Furthermore, it is asymptotically optimal, computationally inexpensive, model-free, and can adapt to changing traffic conditions. Simulations are provided to compare the effectiveness of the proposed algorithm with other known resource-sharing policies such as complete sharing and reservation policies

In this paper we propose an alternative approach for finding a near-optimal call admission policy that prioritizes handoff requests over new calls in a generic mobile cellular network. The performance measure is formed as a weighted linear function of new call and handoff call blocking probabilities. The problem is formulated as a semi-Markov decision process with average cost criterion. Then, a simulation-based learning algorithm based on temporal difference methodology is used to determine a near-optimal control policy online from interaction with the network without a priori knowledge or estimation of the dynamical model of the network. Simulations are provided to compare the effectiveness of the proposed algorithm with two well-known resource-sharing policies: complete sharing and reservation policies (guard threshold). The learning algorithm adapts to traffic variations and this paper shows that it also gives very close blocking probabilities to the optimal guard threshold approach.

In the next generation cellular mobile multimedia networks, a resource allocation policy, which prioritizes handoff requests over new calls while making efficient use of the network resources, will be an essential component for successful operation. In this paper we develop a new handoff prioritized scheme which adapts the allocation policy according to the current traffic conditions. The goal is to minimize the new call blocking while keeping the handoff failures close to a targeted objective. This problem is formulated as a constrained semi-Markov decision process (SMDP) with average cost criterion. A simulation-based learning algorithm is developed to determine a control policy from direct interaction with the network without a priori knowledge of the network dynamics or traffic. Extensive simulations test the effectiveness of the algorithm under a variety of traffic conditions. Comparisons with other resource allocation policies, such as complete sharing and channel reservation, are presented.