Framework design for privacy preserving machine learning in a 6G vehicular network

Date

2026-9

Author

Uçar, Alper Umut

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

71
views

0
downloads

Cite This

The transition to sixth-generation (6G) telecommunications introduces a paradigm shift in vehicular networks, enabling hyper-connected autonomous driving through advanced machine learning. However, training these models requires vast amounts of sensitive telemetry data, creating a conflict between data utility and user privacy. This thesis addresses this trade-off by proposing a dual-framework architecture whose feasibility rests on three concrete 6G capabilities: native edge computing at roadside units, which hosts the cryptographic aggregation stages close to the vehicles; network slicing, which reserves dedicated resources for cryptographically intensive traffic; and terabit-class bandwidth, which makes the ciphertext expansion of public-key primitives negligible in transmission time. First, we introduce a soft privacy framework, leveraging ciphertext-policy attribute-based encryption and proxy re-encryption to achieve ultra-low latency suitable for safety-critical applications like collision avoidance. Second, we present a hard privacy framework, utilizing threshold Paillier homomorphic encryption, Shamir-based multi-party computation, and non-interactive zero-knowledge range proofs to establish a zero-trust environment for long-term statistical analysis. Both frameworks are restricted to linear learning models, since they rely on inner-product functional encryption and the additive homomorphism of Paillier. We provide a security analysis that characterizes the collusion boundary of each framework (collusion between a roadside unit and the central authority collapses the soft privacy guarantee, whereas the hard privacy framework tolerates any coalition below the secret-sharing threshold) and qualifies the extent to which range proofs mitigate data poisoning. The security of both frameworks is computational, resting on the decisional composite residuosity, RSA, and discrete logarithm assumptions; information-theoretic guarantees are confined to the secret-sharing layer. Furthermore, a theoretical complexity analysis and network simulations built on a C implementation compare the two frameworks and confirm that while soft privacy optimizes for real-time actuation, hard privacy provides stronger confidentiality at the cost of higher latency, offering a roadmap for privacy-preserving 6G vehicular architecture.

Subject Keywords

6G Vehicular Networks, Privacy-Preserving Machine Learning, Functional Encryption, Homomorphic Encryption, Multi-Party Computation, Zero-Knowledge Proofs, Attribute-Based Encryption, Age of Information

URI

https://hdl.handle.net/11511/120713

Collections

Graduate School of Applied Mathematics, Thesis