Interference-Aware Optimization of Energy-Efficient Relay-Assisted Collaborative CNN Execution
(2024) 2023 IEEE Virtual Conference on Communications (VCC) (28.November.2023)
Files
Interference-AwareOptimizationofEnergy-EfficientRelay-AssistedCollaborativeCNNExecution.pdf
Open Access - Adobe PDF
- 586.51 KB
Details
- Authors
Kilcioglu, Emre 
Stupia, Ivan
- Abstract
- Collaborative execution of convolutional neural networks (CNNs) holds significant promise for enhancing the performance of latency-critical artificial intelligence applications on resource-constrained devices (RCDs). Edge intelligence advocates for shifting computation-intensive tasks to the edge network to improve latency and system reliability. This paper presents an interference-aware energy-efficient architecture for relay-assisted collaborative CNN execution, integrating data and model parallelism, and addressing interference impact on communication and computation energy consumption. With the predefined signalto- interference-noise ratio (SINR) target, the study investigates the trade-off between communication and computation and formulates a convex optimization problem for energy optimization. The proposed architecture is evaluated through simulations, demonstrating its superiority compared to several benchmark scenarios and shedding light on the role of the SINR target in shaping the trade-off between communication and computation.
- Affiliations
Citations
Kilcioglu, E., Stupia, I., & Vandendorpe, L. (2024). Interference-Aware Optimization of Energy-Efficient Relay-Assisted Collaborative CNN Execution. 2023 IEEE Virtual Conference on Communications (VCC), pp. 145-150. https://doi.org/10.1109/VCC60689.2023.10474727