Research Article
Temporal Weighted Averaging for Asynchronous Federated Intrusion Detection Systems
Algorithm
1Temporal weighted averaging.
| (1) | : number of client devices broadcast in each round | | (2) | nnew: number of clients received in each round | | (3) | : threshold server waiting time | | (4) | t: number of timesteps | | (5) | : timestep in which the weight update originates | | (6) | C: initial threshold constant based on observation of training data | | (7) | P(e): expected ratio of devices. | | (8) | P(r): ratio of received devices. | | 9 | procedure Server | | 10 | | | 11 | : server model initialization | | 12 | , B, n: initialize hyperparameters | | 13 | P(e) 0.8 | | 14 | for to t do | | 15 | Broadcast() | | 16 | Wait() | | 17 | buffer | | 18 | | | 19 | P(r) = | | 20 | | | 21 | | | 22 | end for | | 23 | end procedure | | 24 | procedure Client | | 25 | for to t do | | 26 | while buffer do | | 27 | Wait() | | 28 | end while | | 29 | Train () | | 30 | end for | | 31 | end procedure |
|