Mohamed Abdallah

Cyber-Physical Power Systems (CPPS) combine physical processes with digital technologies to enhance power grid monitoring and control. Nonetheless, this integration introduced cybersecurity vulnerabilities that can bypass conventional Bad Data Detection (BDD) mechanisms and compromise state estimation accuracy. To counter these threats, Moving Target Defense (MTD) strategies employing Distributed Flexible AC Transmission System (D-FACTS) devices have been proposed, dynamically altering system parameters to invalidate the attacker’s knowledge of the system. While few approaches in the literature have been proposed to optimize perturbations introduced by the MTD mechanism, they often rely on computationally intensive iterative optimization techniques, which are challenging for real-time use. To address this gap, this paper proposes a novel approach to optimize the perturbation parameters of D-FACTS devices using Deep Reinforcement Learning (DRL), specifically leveraging the Deep Deterministic Policy Gradient (DDPG) and the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms. The DRL-based method efficiently navigates the continuous action space of D-FACTS devices, enabling real-time adjustments that enhance the detection and identification of FDI attacks without the computational overhead of iterative methods. The proposed approach is validated on simulated IEEE 14-bus and IEEE 57-bus systems, which achieve up to 28% higher attack detection probability compared to Max-Rank MTD and a comparable performance to state-of-the-art iterative robust MTD, while providing a substantial reduction in computational time—up to several orders of magnitude—and maintaining high detection rates across varying attack intensities and perturbation configurations.

As large language models (LLMs) become deeply embedded in daily life, the urgent need for safer moderation systems that distinguish between naive and harmful requests while upholding appropriate censorship boundaries has never been greater. While existing LLMs can detect dangerous or unsafe content, they often struggle with nuanced cases such as implicit offensiveness, subtle gender and racial biases, and jailbreak prompts, due to the subjective and context-dependent nature of these issues. Furthermore, their heavy reliance on training data can reinforce societal biases, resulting in inconsistent and ethically problematic outputs. To address these challenges, we introduce GuardEval, a unified multi-perspective benchmark dataset designed for both training and evaluation, containing 106 fine-grained categories spanning human emotions, offensive and hateful language, gender and racial bias, and broader safety concerns. We also present GemmaGuard (GGuard), a Quantized Low-Rank Adaptation (QLoRA), fine-tuned version of Gemma3-12B trained on GuardEval, to assess content moderation with fine-grained labels. Our evaluation shows that GGuard achieves a macro F1 score of 0.832, substantially outperforming leading moderation models, including OpenAI Moderator (0.64) and Llama Guard (0.61). We show that multi-perspective, human-centered safety benchmarks are critical for mitigating inconsistent moderation decisions. GuardEval and GGuard together demonstrate that diverse, representative data materially improve safety, and adversarial robustness on complex, borderline cases.

The widespread adoption of electric vehicles (EVs) has led to a significant increase in the deployment of charging infrastructures, making them critical components of modern power grids. However, integrating advanced communication technologies into these infrastructures has also enabled sophisticated cyber threats. Although existing research has proposed various solutions, many approaches do not sufficiently address the challenges posed by advanced attack methodologies powered by deep reinforcement learning (DRL), which can adapt their strategies over time, coordinate actions across multiple EVs, and remain stealthy under protocol-compliant constraints. In this paper, we present a novel Multi-Layer Adversarial DRL (MLADRL) framework designed to tackle these challenges. Our approach first utilizes DRL to develop more destructive and coordinated attack vectors, where multiple EVs launch attacks, going beyond the single-EV attacks typically seen in the literature. We combine these DRL-generated attack samples with benign samples and handcrafted attacks to build robust datasets. Based on these datasets, we then design a DRL-based Intrusion Detection System (IDS) specifically designed to detect these advanced cyber threats. We evaluate the effectiveness of our framework across multiple attack scenarios, including varying proportions of malicious EVs and different levels of attack complexity. The results demonstrate that our MLADRL framework significantly outperforms traditional detection methods, maintaining high detection accuracy even against more destructive and complex attack strategies. The proposed DRL-based IDS, trained on the generated adversarial datasets, achieves remarkable performance. Compared to baseline methods, our DRL-based IDS attains 99.4% and 99.9% accuracy, 99.4% and 99.9% precision, and 99.6% and 99.9% recall (F1-scores: 99.6% and 99.9%) for LSTM-based and Transformer-based models, respectively.

Large Language Models (LLMs) have revolutionized the field of natural language processing. However, they exhibit some limitations, including a lack of reliability and transparency: they may hallucinate and fail to provide sources that support the generated output. Retrieval-Augmented Generation (RAG) was introduced to address such limitations in LLMs. One popular implementation, RAG-as-a-Service (RaaS), has shortcomings that hinder its adoption and accessibility. For instance, RaaS pricing is based on the number of submitted prompts, without considering whether the prompts are enriched by relevant chunks, i.e., text segments retrieved from a vector database, or the quality of the utilized chunks (i.e., their degree of relevance). This results in an opaque and less cost-effective payment model. We propose Chunk-as-a-Service (CaaS) as a transparent and cost-effective alternative. CaaS includes two variants: Open-Budget CaaS (OB-CaaS) and Limited-Budget CaaS (LB-CaaS), which is enabled by our “Utility-Cost Online Selection Algorithm (UCOSA)”. UCOSA further extends the cost-effectiveness and the accessibility of the OB-CaaS variant by enriching, in an online manner, a subset of the submitted prompts based on budget constraints and utility-cost tradeoff. Our experiments demonstrate the efficacy of the proposed UCOSA compared to both offline and relevance-greedy selection baselines. In terms of the performance metric—the number of enriched prompts (NEP) multiplied by the Average Relevance (AR)—UCOSA outperforms random selection by approximately 52% and achieves around 75% of the performance of offline selection methods. Additionally, in terms of budget utilization, LB-CaaS and OB-CaaS achieve higher performance-to-budget ratios of 140% and 86%, respectively, compared to RaaS, indicating their superior efficiency.