KrumSimulation#

KrumSimulation.

Reference:: Peva Blanchard, El Mahdi El Mhamdi, Rachid Guerraoui, and Julien Stainer. “Machine learning with adversaries: Byzantine tolerant gradient descent.” In Advances in Neural Information Processing Systems 30 (NIPS 2017).

One KrumSimulation instance = one (aggregator, attack, dataset, model) configuration run over multiple synchronous rounds with no learning rate decay.

class krum.simulations.centralised.krum_nips_2017.KrumSimulation(**kwargs: Any)[source]#

Bases: CentralisedSimulation

Distributed SGD simulation.

Compared to the ICML 2018 HiddenVulnerabilitySimulation, this variant:

Uses a fixed learning rate (no scheduler; lr_decay=None, the default inherited from CentralisedSimulation).
Reports misclassification error and cross-entropy loss on the test set.

evaluate() → tuple[float, float][source]#

Evaluate the model on the test set.

Returns:: Tuple of `` (test_loss, test_accuracy)

evaluate_train() → float[source]#

Evaluate the model on the training set.

Returns:: Training loss.

Note

Reads the full training set in a single no_grad pass (self._full_loader batches the concatenation of every worker’s dataset at once). OK for the small datasets used by the bundled experiments (Spambase ≈ 4.5k rows); for larger datasets, sample a fixed-size subset instead.