tff.learning.secure_aggregator

Creates secure aggregator with adaptive zeroing and clipping.

Zeroes out extremely large values for robustness to data corruption on clients, clips to moderately high norm for robustness to outliers. After weighting in mean, the weighted values are summed using cryptographic protocol ensuring that the server cannot see individual updates until sufficient number of updates have been added together. For details, see Bonawitz et al. (2017) https://dl.acm.org/doi/abs/10.1145/3133956.3133982. In TFF, this is realized using the tff.federated_secure_sum_bitwidth operator.

zeroing Whether to enable adaptive zeroing for data corruption mitigation.
clipping Whether to enable adaptive clipping in the L2 norm for robustness. Note this clipping is performed prior to the per-coordinate clipping required for secure aggregation.
weighted Whether the mean is weighted (vs. unweighted).

A tff.aggregators.AggregationFactory.