ProcessIntermediate
In Algorithm 17.3, what is the condition for adopting the offline-trained model directly as the online authentication model, and what happens if the condition is not met?
In Algorithm 17.3, the ECN adopts the offline-trained model directly as the online authentication model when the number of online nodes equals the number of offline nodes (Numonline == Numoffline), after collecting at least δ CSIs per node; it then sets FPHY = 1. If the condition is not met, the ECN fine-tunes the parameters of the last layer of the model using the collected CSIs while freezing all other parameters, updates the online authentication model, and sets FPHY = 1.
Key points
- The direct adoption condition is Numonline == Numoffline.
- Before checking this condition, the ECN must have collected at least δ CSIs per node.
- If the condition is met, the offline-trained TP-Net model is used unchanged as the online PHY-layer authentication model and FPHY is set to 1.
- If the numbers differ, the ECN fine-tunes only the last layer parameters using the collected CSIs, keeping the other parameters frozen.
- After fine-tuning, the online authentication model is updated and FPHY is set to 1.
Source:AI for Cybersecurity_ Research and Practice· AI-powered Physical Layer Security in Industrial Wireless Networks· p. 524–532
AI for Cybersecurity_ Research and Practice
Unknown
John Wiley & Sons, Inc.