Hybrid Classical-Quantum AI Approach for Detecting Cyberattacks in Vehicles

Details

• Creators:
  Li, Shaozhi ; Tewari, Sumanta ; Wang, Yao ; Chowdhury, Mashrur ; Salek, M Sabbir ; Aggarwal, Vaneet ; Ukkusuri, Satish ; Comert, Gurcan
• Corporate Creators:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC)
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Adversarial Attack Artificial Intelligence Computer Security Cyberattacks Hybrid Vehicles Machine Learning Neural Networks Quantum Activation Function Quantum Mechanics Weight Weight-constrained Neural Network

  More +
• Publication/ Report Number:
  14
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report, 01/01/2024 - 05/31/2025
• Corporate Publisher:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC)
• Abstract:
  Machine learning has become central to self-driving vehicles, enabling tasks such as object detection and anomaly recognition. Yet classical methods often face challenges of high complexity, large parameter counts, and limited training efficiency. To address these, we propose quantum and quantum-inspired machine learning approaches that enhance both efficiency and resilience. We first develop a quantum-inspired weight-constrained neural network that requires substantially fewer parameters than classical counterparts, reducing energy costs for training and inference. Combined with a novel dropout-based defense, this model achieves robustness under strong adversarial attacks, with an average accuracy reduction of about 20%. We also explore hybrid quantum-classical convolutional neural networks using angle encoding and quantum activation functions, demonstrating faster training than classical convolutional neural networks. Motivated by these findings, we introduce quantum-inspired activation functions for classical models, which consistently lower training steps compared to standard Tanh. To capture long-range dependencies, we present the Quantum Non-Local Neural Network (QNL-Net), which integrates classical dimensionality reduction with quantum circuits. QNL-Net achieves state-of-the-art binary classification on MNIST and CIFAR-10 while requiring fewer qubits. Beyond supervised learning, we propose the Quantum Natural Policy Gradient (QNPG) algorithm, achieving improved sample complexity O(ϵ-1.5), surpassing the classical lower bound of O(ϵ-2). Finally, we formulate the Vehicle Routing Problem with Arc Interdiction as a QUBO and demonstrate solutions on D-Wave quantum annealers, showing near-optimal performance on small instances. These results highlight the potential of quantum methods to reduce complexity, accelerate training, and strengthen resilience, paving the way for safer and more efficient AI-driven vehicles.
• Format:
  PDF
• Funding:
  69A3552344812 and 69A3552348317
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:47d7d5c73059c1506aea448f3dc1fc2a3dcf6c7300b91ccbca95bdd531ab865b0b4c7540df9c5a882fc1f4702f168ad7e5fe77677c8b3bee37de43957943a19b
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/89833/dot_89833_DS1.pdf
• File Type:
  [PDF - 1.59 MB ]