Yasmeen Alslman

With the rapid improvement of generative adversarial networks (GANs) in the landscape of data augmentation and generation, researchers have started to explore the use of GANs in other fields, such as adversarial attack generation. Using GANs has improved the field of adversarial attacks, particularly against intrusion detection systems (IDSs). GAN-based adversarial attacks are considered sophisticated attacks that can significantly compromise the security of IDSs. Addressing GAN-based adversarial challenges can ensure the effectiveness and severity of such attacks, which, in turn, lead to the development of strong defense strategies. Few surveys covered the effects and challenges of adversarial attacks against IDSs, focusing on gradient-based adversarial attacks. Recently, many research papers have examined GAN-based adversarial attacks. This paper provides a systematic and comprehensive review of recent research on GAN-based adversarial attacks targeting IDSs, complemented by a meta-analysis guided by critical review questions. Through case studies, we empirically demonstrate the vulnerabilities of binary and multi-class IDS models under GAN-based attacks, examining the influence of feature selection. We are the first to employ advanced GAN variants to craft novel adversarial attacks. Using Fréchet inception distance and Hellinger distance, we assess IDS performance and the realism of generated samples, revealing key system weaknesses. The paper concludes by summarizing the current state of knowledge, outlining major challenges, proposing future research directions, and identifying the most severe GAN architecture and the one that generates the most realistic adversarial samples.

This paper investigates the capabilities of large language models (LLMs) in formulating and solving decision-making problems using mathematical programming. We first conduct a systematic review and meta-analysis of recent literature to assess how well LLMs understand, structure, and solve optimization problems across domains. The analysis is guided by critical review questions focusing on learning approaches, dataset designs, evaluation metrics, and prompting strategies. Our systematic evidence is complemented by targeted experiments designed to evaluate the performance of state-of-the-art LLMs in automatically generating optimization models for problems in computer networks. Using a newly constructed dataset, we apply three prompting strategies: Act-as-expert, chain-of-thought, and self-consistency, and evaluate the obtained outputs based on optimality gap, token-level F1 score, and compilation accuracy. Results show promising progress in LLMs’ ability to parse natural language and represent symbolic formulations, but also reveal key limitations in accuracy, scalability, and interpretability. These empirical gaps motivate several future research directions, including structured datasets, domain-specific fine-tuning, hybrid neuro-symbolic approaches, modular multi-agent architectures, and dynamic retrieval via chain-of-RAGs. This paper contributes a structured roadmap for advancing LLM capabilities in mathematical programming.

Generative adversarial networks (GANs) have advanced rapidly in data augmentation and generation, and researchers have been exploring their applications in other areas, including adversarial attack generation. GANs have significantly improved the field of adversarial attacks, especially against intrusion detection systems (IDSs). These highly sophisticated GAN-based attacks pose a significant threat to IDS security. Addressing the challenges posed by GAN-based adversarial attacks is crucial for assessing their impact and developing robust defense mechanisms. In this paper, we propose new adversarial attack generation techniques derived from advanced GAN architectures. The effectiveness of these attacks, which are based on sophisticated types of GANs, is evaluated against three multi-class IDSs designed for 5G networks. Comprehensive experiments are conducted to evaluate the effectiveness of each GAN in exploiting IDS vulnerabilities and examine the transferability of adversarial attacks across different IDSs, considering the quality of the adversarial samples generated. Exploring these sophisticated GANs for adversarial attack generation enables us to develop a new post-adversarial recovery mechanism based on reconstructing the adversarial samples. Moreover, to thoroughly assess the capabilities of the proposed techniques, new evaluation metrics are introduced to facilitate a comprehensive analysis of the system’s vulnerabilities. Our results show that the proposed GAN-based adversarial attacks can significantly impact IDSs by achieving a high attack success rate (ASR) and drastically reducing accuracy, recall, precision, and F1 score. However, the proposed post-adversarial recovery process effectively restores the IDSs’ performance while significantly reducing the ASR.

Due to the increased attacks on different applications, data security has become crucial. Many modes can be used to operate the advanced encryption standard (AES), some of which provide integrity, and some outperform other modes in security and simplicity. In this paper, the chain block cipher (CBC) mode has been modified to provide more security to the encrypted data by making it robust against the bit-flipping attack and adding an integrity approach using the keyedhash function. In addition, using the keyd-hash function increases the number of keys needed in CBC-AES to two keys, and this can make the proposed model more secure against bruteforce attacks and Grover’s quantum search algorithm.

Due to the increased attacks on different applications, data security has become crucial. Many modes can be used to operate the advanced encryption standard (AES), some of which provide integrity, and some outperform other modes in security and simplicity. In this paper, the chain block cipher (CBC) mode has been modified to provide more security to the encrypted data by making it robust against the bit-flipping attack and adding an integrity approach using the keyedhash function. In addition, using the keyd-hash function increases the number of keys needed in CBC-AES to two keys, and this can make the proposed model more secure against bruteforce attacks and Grover’s quantum search algorithm.

Recently, medical image encryption has gained special attention due to the nature and sensitivity of medical data and the lack of effective image encryption using innovative encryption techniques. Several encryption schemes have been recommended and developed in an attempt to improve medical image encryption. The majority of these studies rely on conventional encryption techniques. However, such improvements have come with increased computational complexity and slower processing for encryption and decryption processes. Alternatively, the engagement of intelligent models such as deep learning along with encryption schemes exhibited more effective outcomes, especially when used with digital images. This paper aims to reduce and change the transferred data between interested parties and overcome the problem of building negative conclusions from encrypted medical images. In order to do so, the target was to transfer from the domain of encrypting an image to encrypting features of an image, which are extracted as float number values. Therefore, we propose a deep learning-based image encryption scheme using the autoencoder (AE) technique and the advanced encryption standard (AES). Specifically, the proposed encryption scheme is supposed to encrypt the digest of the medical image prepared by the encoder from the autoencoder model on the encryption side. On the decryption side, the analogous decoder from the auto-decoder is used after decrypting the carried data. The autoencoder was used to enhance the quality of corrupted medical images with different types of noise. In addition, we investigated the scores of structure similarity (SSIM) and mean square error (MSE) for the proposed model by applying four different types of noise: salt and pepper, speckle, Poisson, and Gaussian. It has been noticed that for all types of noise added, the decoder reduced this noise in the resulting images. Finally, the performance evaluation demonstrated that our proposed system improved the encryption/decryption overhead by 50–75% over other existing models.

Privacy-preserving of medical information (such as medical records and images) is an essential right for patients to ensure security against undesired access parties. This right is typically protected by law through firm regulations set by healthcare authorities. However, sensitive-private data usually requires the application of further security and privacy mechanisms such as encipherment (encryption) techniques. ’Medical images’ is one such example of highly demanding security and privacy standards. This is due to the quality and nature of the information carried among these images, which are usually sensitive-private information with few features and tonal variety. Hence, several state-of-the-art encryption mechanisms for medical images have been proposed and developed; however, only a few were efficient and promising. This paper presents a hybrid crypto-algorithm, MID-Crypt, to secure the medical image communicated between medical laboratories and doctors’ accounts. MID-Crypt is designed to efficiently hide medical image features and provide high-security standards. Specifically, MID-Crypt uses a mix of Elliptic-curve Diffie–Hellman (ECDH) for image masking and Advanced Encryption Standard (AES) with updatable keys for image encryption. Besides, a key management module is used to organize the public and private keys, the patient’s digital signature provides authenticity, and integrity is guaranteed by using the Merkle tree. Also, we evaluated our proposed algorithm in terms of several performance indicators including, peak signal-to-noise ratio (PSNR) analysis, correlation analysis, entropy analysis, histogram analysis, and timing analysis. Consequently, our empirical results revealed the superiority of MID-Crypt scoring the best performance values for PSNR, correlation, entropy, and encryption overhead. Finally, we compared the security measures for the MID-Crypt algorithm with other studies, the comparison revealed the distinguishable security against several common attacks such as side-channel attacks (SCA), differential attacks, man-in-the-middle attacks (MITM), and algebraic attacks.

Physical documents have many challenges, due to being difficult to exchange and the potential of being lost and sabotaged. Digital documents seem to address some of these challenges, but because of the use of the Internet and reliance on third parties to share and preserve such documents, the privacy and integrity of the digital documents became difficult to maintain. The blockchain technology that Satoshi Nakamoto has proposed for financial transactions can be used to solve the problems of trust in centralized systems and the subversion and alteration thatthese transactions may be exposed to. Blockchain-based systems have shown their effectiveness and ability to solve most security problems related to exchanging systems' transactions. Hence,A decentralized system for exchanging digital documents, using Ethereum blockchain and InterPlanetary File System (IPFS), has been proposed to overcome the problems of using trusted third-parties in exchanging digital documents. The proposed system provides immutable and irreversible logs of the process of sending and receiving documents. In this framework, all registered users can encrypt the documents using the Advanced Encryption Standard (AES) Algorithm, send, and receive documents using the Ethereum Blockchain's smart contract. The framework was simulated locally using Ganache Blockchain and tested on the official Ethereum test network Ropsten, and the experimental results showed the feasibility of the proposed scheme. The smart contract used in the Ethereum blockchain was written in the Solidity programming language, and the website that was needed to interact with the blockchain was built using React and other different javascript libraries like web3. Additionally, Metamask wallet was used to create the Ethereum accounts which are required to connect and communicate with the Smart Contract that is implemented in the Ethereum blockchain. Finally, the model's results indicated that the model may cost an average of 0.0552$ per transaction, and this shows that the system is affordable. Also, the model has satisfied the three main security requirements which are confidentiality, integrity, and availability by using the suitable countermeasure for each one of them.