Research Article
Fed-DNN-Debugger: Automatically Debugging Deep Neural Network Models in Federated Learning
| Require: total number of users , global training rounds , local training rounds , a buggy model , training dataset , test dataset , target accuracy ; | | Ensure: well debugged model ; | | Server-orchestrated training://Server side | (1) | for each round from 0 to do | (2) | (test model using ) | (3) | ifthen | (4) | for each user from 0 to in parallel do | (5) | ClientDebugger(n, ) | (6) | end for | (7) | | (8) | else | (9) | return | (10) | end if | (11) | end for | (12) | returnClientDebugger://Client side | (13) | | (14) | //Nonintrusive Metadata Capture | (15) | //Automated Neural Network Model Debugging | (16) | for each round from 0 to do | (17) | ( split into size batches) | (18) | | (19) | end for | (20) | return |
|