Enhancing IoT Anomaly Detection Performance for Federated Learning

Student: Brett Weinger (Stony Brook University)
Supervisor: Jinoh Kim (Lawrence Berkeley National Laboratory)

Abstract: While federated learning has gained great attention for mobile computing with the benefits of scalable cooperative learning and privacy protection capabilities, there still exist a great deal of technical challenges to make it practically deployable. For instance, distribution of the training process to a myriad of devices limits the classification performance of deep neural networks, often showing significantly degraded metrics compared to centralized learning. We propose that data augmentation can be used to ensure that client nodes have a sufficiently balanced set of examples for each class and will contribute more optimal updates to the global model. We show experimental results that employ statistical data generation techniques including random sampling and k-nearest neighbor algorithms. Our results indicate that random sampling of minority class data yields significant improvements for our IoT anomaly detection use case with manageable learning complexity.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: PDF


Back to Poster Archive Listing