Resource-Aware Secure Computation Offloading

Computation offloading enables resource-limited users to delegate resource-intensive processing tasks to more powerful edge devices (workers). In doing so, a major concern is to protect the privacy of sensitive data against untrusted devices. In this work, we propose a secure computation offloading framework to address this challenge. To enable adaptability to resource heterogeneities across the workers, we then propose a hierarchical combinatorial neural multi-armed bandit mechanism for secure computation resource allocation. Our framework provides information-theoretic privacy guarantees for both the sensitive user data and computation results, while minimizing system latency of secure computations in the presence of heterogeneous resource availability. Our experiment results show that the proposed framework achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3 \times$</tex> faster task completion time compared to conventional secure offloading baselines and is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$13 \times$</tex> faster than local processing at the user.