Van-Linh Nguyen

This paper aims to propose a novel machine learning (ML) approach incorporating Homomorphic Encryption (HE) to address privacy limitations in Unmanned Aerial Vehicles (UAV)-based face detection. Due to challenges related to distance, altitude, and face orientation, high-resolution imagery and sophisticated neural networks enable accurate face recognition in dynamic environments. However, privacy concerns arise from the extensive surveillance capabilities of UAVs. To resolve this issue, we propose a novel framework that integrates HE with advanced neural networks to secure facial data throughout the inference phase. This method ensures that facial data remains secure with minimal impact on detection accuracy. Specifically, the proposed system leverages the Cheon-Kim-Kim-Song (CKKS) scheme to perform computations directly on encrypted data, optimizing computational efficiency and security. Furthermore, we develop an effective data encoding method specifically designed to preprocess the raw facial data into CKKS form in a Single-Instruction-Multiple-Data (SIMD) manner. Building on this, we design a secure inference algorithm to compute on ciphertext without needing decryption. This approach not only protects data privacy during the processing of facial data but also enhances the efficiency of UAV-based face detection systems. Experimental results demonstrate that our method effectively balances privacy protection and detection performance, making it a viable solution for UAV-based secure face detection. Significantly, our approach (while maintaining data confidentially with HE encryption) can still achieve an accuracy of less than 1% compared to the benchmark without using encryption.

Quantum computing has been a major priority for several nations and prominent institutions in their pursuit of a transformative breakthrough in the fields of computation and encryption. By using the principles of quantum mechanics, particularly quantum superposition and entanglement, quantum computing and quantum machine learning (QML) have the potential to enhance artificial intelligence (AI) and achieve quantum supremacy with unprecedented computational power. However, despite its exceptional learning capabilities, QML-based applications face several emerging security threats. Unlike previous studies focused on classical quantum cryptography and secure quantum communications, this work investigates adversarial risks in QML-assisted network functions and digital twin applications. Specifically, we highlight vulnerabilities such as quantum kernel poisoning, backdoor attacks, and adversarial noise. Key findings reveal that adversaries can intercept quantum states in transit, manipulate parameterized quantum circuits (PQCs), and exploit variational quantum algorithms (VQAs) through adversarial qubit perturbations. These attacks can mislead QML-based optimization processes, leading to incorrect digital twin predictions, faulty resource allocation, or disruptions in QML-aided network functions. To mitigate these risks, defense strategies such as quantum-safe cryptography, data sanitization, adversarial training, defensive distillation, and gradient masking in quantum circuit design can be employed. However, the key issue is the absence of robust security solutions for real-world deployment. Future research should examine the trade-off between adversarial robustness and generative learning performance. Key areas include quantum state discrimination, secure quantum federated learning, quantum decoherence control, and secure quantum semantic communications for real-world deployment.

The advent of open radio access networks (O-RAN) has introduced intelligent, flexible, and multi-vendor network ecosystems. While O-RAN’s open interfaces and artificial intelligence (AI)-driven solutions offer improved performance, energy efficiency, and resource minimization for green networking, they also expose the system to new security vulnerabilities, particularly adversarial attacks. This paper presents a robust defense approach, termed RADAR, designed to secure deep reinforcement learning (DRL)-powered resource allocation mechanisms in O-RAN. RADAR is a multi-faceted defense framework that integrates adversarial input sanitization, proactive adversarial training, and adapted defensive distillation to counter policy infiltration attacks, gradient-based deceptive loss maximization, and signal perturbation injections into the O-CU via the O-DU in O-RAN. This study evaluates the effectiveness of RADAR not only against a novel attack variant—policy infiltration attack (PIA), which manipulates environmental parameters to disrupt allocation decisions, but also against well-known adversarial techniques such as the fast gradient sign method (FGSM) and projected gradient descent (PGD). Experimental results demonstrate that RADAR achieves significant recovery in user data rates across three network slices: 73.33% for eMBB, 64.71% for mMTC and 52.94% for uRLLC, outperforming the existing standalone approach. The findings highlight RADAR’s effectiveness in mitigating adversarial attack techniques, underscoring its potential to secure AI-driven core functions in intelligent O-RAN.