[CVPR 21] Soteria: Provable Defense Against Privacy Leakage in Federated Learning From Representation Perspective (thecvf.com)
- Observation: the class-wise data reprensentations of each device’s data are embedded in shared local model updates, and such data representations can be inferred to perform model inversion attacks. Therefore, the information can be severely leaked through the model updates.
- 矩阵求导法则与性质 | zdaiot
[ICML 23] Revisiting Weighted Aggregation in Federated Learning with Neural Networks (mlr.press)
- Observation: smaller l1 norms of aggregation weights may be beneficial to generalization.
- Devise FEDLAW, learning the weights and boosting the generalization of global models.