Batch-level aggregated gradients, LoRA adapters, and de-biased label differential privacy let advertisers and platforms train conversion models in a clean room with modest AUC loss and much lower communication cost.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Privacy Preserving Conversion Modeling in Data Clean Room
Batch-level aggregated gradients, LoRA adapters, and de-biased label differential privacy let advertisers and platforms train conversion models in a clean room with modest AUC loss and much lower communication cost.