Privacy drift, the change in membership inference risk during incremental training, correlates strongly with training accuracy across centralized and federated settings.
Large scale incremental learning,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Privacy Drift: Evolving Privacy Concerns in Incremental Learning
Privacy drift, the change in membership inference risk during incremental training, correlates strongly with training accuracy across centralized and federated settings.