A novel place recognition method for large-scale forest scenes

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 126606 - 126606

Published: Jan. 1, 2025

Language: Английский

---

AI Under Attack: Metric-Driven Analysis of Cybersecurity Threats in Deep Learning Models for Healthcare Applications

Algorithms, Journal Year: 2025, Volume and Issue: 18(3), P. 157 - 157

Published: March 10, 2025

Incorporating Artificial Intelligence (AI) in healthcare has transformed disease diagnosis and treatment by offering unprecedented benefits. However, it also revealed critical cybersecurity vulnerabilities Deep Learning (DL) models, which raise significant risks to patient safety their trust AI-driven applications. Existing studies primarily focus on theoretical or specific attack types, leaving a gap understanding the practical implications of multiple scenarios AI. In this paper, we provide comprehensive analysis key vectors, including adversarial attacks, such as gradient-based Fast Gradient Sign Method (FGSM), evasion attacks (perturbation-based), data poisoning, threaten reliability DL with breast cancer detection. We propose Healthcare AI Vulnerability Assessment Algorithm (HAVA) that systematically simulates these calculates Post-Attack Index (PAVI), quantitatively evaluates impacts. Our findings FGSM significantly reduced model accuracy from 97.36% 61.40% (PAVI: 0.385965) 62.28% 0.377193), respectively, demonstrating severe impact performance, but poisoning had milder effect, retaining 89.47% 0.105263). The confusion matrices higher rate false positives than more balanced misclassification patterns observed poisoning. By proposing unified framework for quantifying analyzing post-attack vulnerabilities, research contributes formulating resilient models domains where are important.

Language: Английский

---

Advanced Persistent Threat Detection Using Optimized and Hybrid Deep Learning Approach

Security and Privacy, Journal Year: 2025, Volume and Issue: 8(2)

Published: Feb. 26, 2025

ABSTRACT Advanced persistent threats ( APT ) are a challenging threat in cybersecurity because of their stealth, persistence, and adaptation to evade traditional detection systems. To tackle this issue, we put forward an optimized deep learning approach that combines Convolutional Neural Network—Long Short‐Term Memory CNN ‐ LSTM architecture with the lime mold algorithm SMA for better detection. During training, balances exploration exploitation well, leading faster convergence performance. The ‐optimized was evaluated on Unraveled dataset, benchmark network intrusion detection, 94.3% accuracy precision, recall, F1 scores 92.8%, 93.5%, 93.1%, respectively. Furthermore, model had false positive rate 2% negative 3% thus more able detect. Scalability tests confirmed model's efficiency at handling high traffic, distributed training processing 50,000 records/s reducing time by 35% over single‐node setups. These results show combining novel optimization techniques is effective proposed framework robust, scalable, efficient, it significantly advances real‐time improves resilience critical infrastructures.

Language: Английский

---

A Novel Approach to Automated Cybersecurity Response for Critical Infrastructures Using Graph Neural Networks and Reinforcement Learning

Lecture notes on data engineering and communications technologies, Journal Year: 2025, Volume and Issue: unknown, P. 35 - 47

Published: Jan. 1, 2025

Language: Английский

---

CLDM-MMNNs: Cross-layer defense mechanisms through multi-modal neural networks fusion for end-to-end cybersecurity—Issues, challenges, and future directions

Information Fusion, Journal Year: 2025, Volume and Issue: unknown, P. 103222 - 103222

Published: April 1, 2025

Language: Английский

---

A High-Accuracy Advanced Persistent Threat Detection Model: Integrating Convolutional Neural Networks with Kepler-Optimized Bidirectional Gated Recurrent Units

Electronics, Journal Year: 2025, Volume and Issue: 14(9), P. 1772 - 1772

Published: April 27, 2025

Advanced Persistent Threat (APT) refers to a highly targeted, sophisticated, and prolonged form of cyberattack, typically directed at specific organizations or individuals. The primary objective such attacks is the theft sensitive information disruption critical operations. APT are characterized by their stealth complexity, often resulting in significant economic losses. Furthermore, these may lead intelligence breaches, operational interruptions, even jeopardize national security political stability. Given covert nature extended durations attacks, current detection solutions encounter challenges as high difficulty insufficient accuracy. To address limitations, this paper proposes an innovative high-accuracy attack model, CNN-KOA-BiGRU, which integrates Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BiGRU), Kepler optimization algorithm (KOA). model first utilizes CNN extract spatial features from network traffic data, followed application BiGRU capture temporal dependencies long-term memory, thereby forming comprehensive features. Simultaneously, employed optimize structure, achieving globally optimal feature weights enhancing Additionally, study employs combination sampling techniques, including Synthetic Minority Over-sampling Technique (SMOTE) Tomek links, mitigate classification bias caused dataset imbalance. Evaluation results on CSE-CIC-IDS2018 experimental demonstrate that CNN-KOA-BiGRU achieves superior performance detecting with average accuracy 98.68%. This surpasses existing methods, (93.01%), CNN-BiGRU (97.77%), Graph Network (GCN) (95.96%) same dataset. Specifically, proposed demonstrates improvement 5.67% over CNN, 0.91% CNN-BiGRU, 2.72% GCN. Overall, 3.1% compared methods.

Language: Английский

---

SDN-DDoSNet: A Deep Learning Framework for DDoS Attack Detection in Software-Defined Networks

Published: Dec. 6, 2024

Language: Английский

---