Krum

class torchjd.aggregation.Krum(n_byzantine, n_selected=1)

Aggregator for adversarial federated learning, as defined in Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent.

Parameters:

• n_byzantine (int) – The number of rows of the input matrix that can come from an adversarial source.

• n_selected (int) – The number of selected rows in the context of Multi-Krum. Defaults to 1.

class torchjd.aggregation.KrumWeighting(n_byzantine, n_selected=1)

Weighting giving the weights of Krum.

Parameters:

• n_byzantine (int) – The number of rows of the input matrix that can come from an adversarial source.

• n_selected (int) – The number of selected rows in the context of Multi-Krum. Defaults to 1.